Springe zum Content
UNIGIS
  • Willkommen
  • Studienangebot
    • UNIGIS professional
    • UNIGIS MSc
    • Studiengangswahl
    • Optionale Module
    • Für Unternehmen
    • Demo Modul
  • Rund ums Studieren
    • Online Studieren
    • Qualitätssicherung
    • Masterarbeiten
    • Absolvent*innenstimmen
    • Blog
    • Link des Monats
    • FAQ
  • Netzwerk
    • Club UNIGIS
    • Magazin
    • Stellenmarkt
    • Kooperationen
    • UNIGIS weltweit
  • Über uns
    • Team
    • Fachbereich Geoinformatik
    • Universität Salzburg
  • Kontakt
  • Willkommen
  • Studienangebot
    • UNIGIS professional
    • UNIGIS MSc
    • Studiengangswahl
    • Optionale Module
    • Für Unternehmen
    • Demo Modul
  • Rund ums Studieren
    • Online Studieren
    • Qualitätssicherung
    • Masterarbeiten
    • Absolvent*innenstimmen
    • Blog
    • Link des Monats
    • FAQ
  • Netzwerk
    • Club UNIGIS
    • Magazin
    • Stellenmarkt
    • Kooperationen
    • UNIGIS weltweit
  • Über uns
    • Team
    • Fachbereich Geoinformatik
    • Universität Salzburg
  • Kontakt
English program
Logo der Universität Salzburg
November 11., 2026

UNIGIS Master Thesis: How to Detect Gradual Loss of Tree Vitality Using Semantic Categories Copy

Figure 1: Semantic-based Disturbance Score (left) compared with a satellite image of the study area (right). Map data ©2015 Google.
Black pines in Lower Austria can look healthy, even though they might be slowly losing their vitality. This article shows how semantic categories from Sentinel-2 imagery can help to detect this gradual decline.

How to Detect Gradual Loss of Tree Vitality Using Semantic Categories

Imagine you are walking through a black pine forest. The trees look green and healthy, but some of them may already be losing vitality. Would you notice it? Obvious loss of tree vitality, such as that caused by fire, is immediately visible. The smaller the disturbances are, the more difficult it is to detect them. But could a satellite 786 kilometers above you detect such a gradual loss of tree vitality? This article presents an approach using semantically enriched pixels, where each pixel is described by a semantic category instead of reflectance values.

Why is gradual loss of tree vitality hard to detect?

“Gradual” refers to a slow loss of vitality that develops over several years before it becomes visible. Such hidden damage is caused, for example, by latent pathogens like Diplodia sapinea. The fungus can live in trees unnoticed for a long time [1] and has long-lasting effects on tree vitality [2]. Consequently, a single on-site inspection does not provide useful information. Reliable information would require regular in-situ monitoring over several years. Even analyses based on satellite imagery face challenges such as mixed pixels or the limited availability of field data for validation.

Gradual loss of tree vitality in Austria’s forests

This kind of hidden decline also affects the black pine forests of Lower Austria, where this study is located. The study area covers about 26 km² of pure black pine stands in the Steinfeld region between Wiener Neustadt and Neunkirchen. The black pine copes well with the changing climate in Europe, as it tolerates, for example, drought and extreme temperatures [3]. However, rising temperatures also promote the spread of Diplodia sapinea [4]. This fungus causes Diplodia tip blight and is considered a global threat, especially for Europe [5]. Starting in southern Europe, the fungus has by now reached Estonia  [6], Sweden [4] and Finland [7]. In Austria, too, damage caused by Diplodia sapinea occurs, including the study area, according to the Austrian Research Centre for Forests (BFW). The BFW provided reference data for this study, including maps of tree species [8] and forest disturbances [9].

What are semantically enriched pixels?

Unlike a land cover class such as “forest”, which usually remains forest over the years regardless of its vitality, semantically enriched pixels describe what they look like rather than what they are. Based on its spectral signature, each pixel is assigned a descriptive label, called a semantic category [10]. For example, a pixel may look like “dense vegetation with low NIR (near infrared reflectance)”, “sparse vegetation with high NIR” or “barren land”. These categories are determined by SIAM (Satellite Image Automatic Mapper), using rules instead of training data [11]. For this study, time series of semantically enriched Sentinel-2 images were accessed via the semantic EO data cube Sen2Cube.at [11], developed at the Department of Geoinformatics – Z_GIS, University of Salzburg.

Can these semantic categories reflect gradual loss of tree vitality?

Yes, with the help of a newly developed index and its temporal pattern. Instead of deriving values directly from the spectral signature, this study analyzed which semantic categories occur and how often over several years. The categories were grouped into dense vegetation (DV) and low vegetation (LV). DV includes the semantic categories that are considered typical for dense black pine stands, such as “average dense vegetation with low NIR”, the most frequent category in the study area. LV includes categories such as sparse vegetation, rangeland or barren land. For each pixel, the number of DV and LV observations was counted per quarter. Based on these counts, a new index was developed in this study, the Semantic-based Disturbance Index (SDI). It is calculated as a normalized difference between the two groups:

SDIt = ∑t DV − ∑t LV ∑t DV + ∑t LV

where DV and LV are the number of observations assigned to each group within t, and t is one quarter of a year (e.g. Q2 2018). This formula puts all pixels on the same scale from -1 to 1. A value close to 1 means the pixel was assigned to dense vegetation only, a value close to -1 that the pixel was assigned to low vegetation only. The index was calculated separately for the second and third quarters of each year between 2018 and 2024.

Figure 2: SDI calculation in Sen2Cube.at

Reading the temporal pattern

Semantic enrichment turns reflectance values into categories with meaning, and the index counts how often each pixel falls into them over the years. If a pure black pine pixel shifts from dense towards low vegetation, this may indicate a loss of vitality. A single index value decline does not yet indicate vitality loss. It only becomes meaningful when several temporal patterns occur. These patterns were identified through time series analysis of sample areas. In the time series charts, Q2 and Q3 were not shown one after the other, but as two overlaid lines across the years. This made a typical pattern visible. In undisturbed years, Q2 values were higher than Q3 values. In affected areas, the two lines came closer together or even crossed.

Mapping the patterns across the forest

To make the temporal patterns visible across the whole study area, the patterns had to be translated into a simpler form. Therefore, the Semantic-based Disturbance Code (SDC) was developed, summarizing the temporal pattern of each pixel and representing the index level, trend and convergence / divergence of the quarterly values.

Figure 3: From index time series to code and score

To make interpretation quick and easy, the resulting 18 codes were further grouped into the Semantic-based Disturbance Score (SDS) with five levels – from 1 (no signs of disturbance) to 5 (severe signs).

Figure 4: Semantic-based Disturbance Score across the study area. Map data ©2015 Google.

What does the score tell us?

Low and moderate signs of disturbance are the most frequent scores in the study area. Advanced and severe signs are less frequent, and only about 7% of the pixels show no signs of disturbance at all. Severe disturbances appear in linear patterns within the forest, along its edges, roads and railways. Advanced disturbances show a similar distribution but are more concentrated along the edge of the forest. These patterns seem to suggest human-driven intervention, such as sanitation felling or forest thinning. Low and moderate signs of disturbance, in contrast, appear as large clusters within the forest. This could be consistent with the short-distance spread of Diplodia sapinea, but also with common location conditions, as nearby trees have the same stress conditions. A cluster analysis (Local Moran’s I) indicated that these patterns are spatially structured rather than random.

How reliable is the score?

Due to the lack of reference data collected on the ground, the score was compared with the forest disturbance levels provided by the BFW, which are based on an established method using red-green vegetation index (RGVI) from Sentinel-2 time series [9]. For each pixel, the rank difference was determined. Differences of +/-1rank were accepted, since vitality loss is a continuous process where boundaries are unclear [12] and both methods create artificial ones. About 75% of the pixels matched within this range. The most frequent difference is a score one level higher than the BFW level. This could reflect different sensitivities of the two approaches, as semantic categories are based on more spectral bands. For the remaining pixels outside +/-1 rank difference, the SDS also mostly shows a higher score than the corresponding BFW level. These larger differences mainly occur at forest edges and roads, where mixed pixels might lead to overestimation. However, without ground-collected reference data for validation, the absolute accuracy cannot be reliably estimated.

What’s next?

This study showed that semantic categories can capture changes related to gradual loss of tree vitality, and that an index-based approach makes it possible to quantify the degree of disturbance. The current process is based on manual processing steps. Automating these steps would improve reproducibility and robustness. Furthermore, subjectivity in definitions of thresholds, coding rules and scores could be reduced by using machine learning methods.

A monthly instead of quarterly analysis could show early signs of vitality loss in more detail, especially the stress phase before the first signs of disturbances. Determining the time at which a disturbance is first detected could help to quantify the time lag between detection and on-site inspection. Together with the SDS map, this could support forest management with early warnings. Spatial outliers in the map, such as single disturbed pixels in an undisturbed area, could help to decide where to inspect first.

Are you curious about the details? Then have a look at my master thesis.

I am grateful to my supervisors Martin Sudmanns and Dirk Tiede (Department of Geoinformatics – Z_GIS, University of Salzburg) for their guidance and motivation throughout this master thesis. Many thanks also to the Austrian Research Centre for Forests (BFW) for providing the data that formed the basis of this work, and for their helpful responses to my questions.

Referenzen

  • K. Blumenstein, J. Bußkamp, G. J. Langer, E. J. Langer, and E. Terhonen, “The Diplodia Tip Blight Pathogen Sphaeropsis sapinea Is the Most Common Fungus in Scots Pines’ Mycobiome, Irrespective of Health Status—A Case Study from Germany,” J. Fungi, vol. 7, no. 8, p. 607, Jul. 2021, doi: 10.3390/jof7080607.
  • M. Caballol, M. Ridley, M. Colangelo, C. Valeriano, J. Julio Camarero, and J. Oliva, “Tree mortality caused by Diplodia shoot blight on Pinus sylvestris and other mediterranean pines,” For. Ecol. Manag., vol. 505, p. 119935, Feb. 2022, doi: 10.1016/j.foreco.2021.119935.
  • Z. Vacek, J. Cukor, S. Vacek, J. Gallo, V. Bažant, and A. Zeidler, “Role of black pine (Pinus nigra J. F. Arnold) in European forests modified by climate change,” Eur. J. For. Res., vol. 142, no. 6, pp. 1239–1258, Dec. 2023, doi: 10.1007/s10342-023-01605-5.
  • L. Brodde et al., “Diplodia Tip Blight on Its Way to the North: Drivers of Disease Emergence in Northern Europe,” Front. Plant Sci., vol. 9, p. 1818, Jan. 2019, doi: 10.3389/fpls.2018.01818.
  • M. J. Wingfield, B. Slippers, I. Barnes, T. A. Duong, and B. D. Wingfield, “The Pine Pathogen Diplodia sapinea: Expanding Frontiers,” Curr. For. Rep., vol. 11, no. 1, p. 2, Dec. 2024, doi: 10.1007/s40725-024-00236-2.
  • K. Adamson, D. Klavina, R. Drenkhan, T. Gaitnieks, and M. Hanso, “Diplodia sapinea is colonizing the native Scots pine (Pinus sylvestris) in the northern Baltics,” Eur. J. Plant Pathol., vol. 143, no. 2, pp. 343–350, Oct. 2015, doi: 10.1007/s10658-015-0686-8.
  • M. M. Müller, J. Hantula, M. Wingfield, and R. Drenkhan, “Diplodia sapinea found on Scots pine in Finland,” For. Pathol., vol. 49, no. 1, p. e12483, Feb. 2019, doi: 10.1111/efp.12483.
  • T. Schadauer et al., “Evaluating Tree Species Mapping: Probability Sampling Validation of Pure and Mixed Species Classes Using Convolutional Neural Networks and Sentinel-2 Time Series,” Remote Sens., vol. 16, no. 16, p. 2887, Aug. 2024, doi: 10.3390/rs16162887.
  • M. Löw and T. Koukal, “Phenology Modelling and Forest Disturbance Mapping with Sentinel-2 Time Series in Austria,” Remote Sens., vol. 12, no. 24, p. 4191, Dec. 2020, doi: 10.3390/rs12244191.
  • A. Baraldi, L. Durieux, D. Simonetti, G. Conchedda, F. Holecz, and P. Blonda, “Automatic Spectral-Rule-Based Preliminary Classification of Radiometrically Calibrated SPOT-4/-5/IRS, AVHRR/MSG, AATSR, IKONOS/QuickBird/OrbView/GeoEye, and DMC/SPOT-1/-2 Imagery—Part I: System Design and Implementation,” IEEE Trans. Geosci. Remote Sens., vol. 48, no. 3, pp. 1299–1325, Mar. 2010, doi: 10.1109/TGRS.2009.2032457.
  • M. Sudmanns, H. Augustin, L. Van Der Meer, A. Baraldi, and D. Tiede, “The Austrian Semantic EO Data Cube Infrastructure,” Remote Sens., vol. 13, no. 23, p. 4807, Nov. 2021, doi: 10.3390/rs13234807.
  • R. G. Congalton and K. Green, Assessing the accuracy of remotely sensed data: principles and practices, 2. ed. Boca Raton, Fla.: CRC Press, 2009.

Claudia Münch

Claudia Münch has almost 27 years of experience at Deutsche Börse and Börse Stuttgart. At Deutsche Börse, she worked at StatistiX Information Management and the index business. Today, she is Product Owner of SID (System for Information Distribution) at Börse Stuttgart. Data has always been part of her work, with a focus on creating information and gaining new insights from it. Her interest in GIS started on walks in the forest, where she saw the effects of drought. With her Master of Science in Geographical Information Science & Systems from UNIGIS Salzburg (2026), she added the spatial component to her data expertise.

Zusätzliche Links

Master Thesis
LinkedIn
Sen2Cube
BFW
Zurück zum Blog
Zurück nach oben

Weitere Beiträge

März 1., 2023

How to Teach in the Age of ChatGPT A Survival Guide

Beitrag anzeigen
März 16., 2023

R Markdown Introduced in the R Module

Beitrag anzeigen
April 25., 2023

UNIGIS Optionale Module (Tag 2023): Gestaltungsspielraum im UNIGIS Studium

Beitrag anzeigen
Mai 11., 2023

GIS goes Gaming

Beitrag anzeigen
August 2., 2023

UNIGIS at GI_Salzburg 2023 – A Throwback

Beitrag anzeigen
August 4., 2023

Esri UC – The Geeky Wonderland You Never Knew You Needed!

Beitrag anzeigen

© 2026 by Z_GIS

  • Impressum
  • Barrierefreiheit