Drone remote sensing for forest health monitoring

Abstract: Forest health is critical for maintaining the multitude of ecosystem ser-
vices that forests provide, including carbon sequestration, biodiversity
conservation, water regulation, and soil conservation. The ability of
forests to perform these functions is directly linked to their health con-
dition, making forest health monitoring essential for sustainable forest
management and climate change mitigation. Traditional methods of
monitoring forest health involve time-consuming and labor-intensive,
sample-based ground observations. While valuable, these methods
are limited in scope and scalability and include highly subjective
evaluations.

This research addresses these limitations by exploring Uncrewed
Aerial Vehicle (UAV)-based forest health monitoring as a viable alter-
native. This dissertation is based on three articles that were prepared
as part of a multi-year remote sensing research project at the Bavarian
State Institute of Forestry (LWF). In the course of the project, 235
Level-1-Monitoring plots of the International Co-operative Program
on Assessment and Monitoring of Air Pollution Effects on Forests (ICP
Forests) were surveyed with UAVs in parallel to the terrestrial inven-
tories in the years 2020– 2022. In this way multispectral aerial images
were collected and merged into a large long-term and cross-temporal
time-series dataset.

The first article scopes the theoretical foundation of the work. With
a review that provides researchers and practitioners with an overview
of previous work related to forest health monitoring and that intro-
duces the latest technology. To achieve this, 99 papers were evaluated,
offering a broad perspective on advancements and methodologies in
the field of UAV-based forest health monitoring. The review identi-
fied research gaps and trends to guide future research efforts and
directions.

Based on this gathered knowledge, the subsequent articles built on
these insights to develop and test innovative UAV-based monitoring
techniques. Consequently, the research presented in the second article
describes the development of an open-source data pipeline that aims
to link UAV data with field data of forest health assessments in a
standardized and streamlined process. This contributed to the semi-
automatic generation of training data for the training of deep learning
models. In a large-scale flight campaign, multispectral UAV data from
235 ICP Forests inventory plots in Bavaria were recorded annually over
the years 2020- 2022. The field data from the same inventory points of
the related years were used as a reference to validate the aerial data.
With the developed pipeline, more than 17,000 training samples of the
five major tree species occurring in Germany including their health
status, two genus classes as well as dead trees could be generated.
In this way, we were able to classify 14 different classes with an
average macro F1-score of 0.61 using the EfficientNet Convolutional
Neural Network (CNN) architecture. The highest class-specific F1
score besides the class of dead trees (0.97) was achieved by the class
of healthy Picea abies (0.80).

Originating from the same database, species-specific gradient-
boosting models were trained. The results, presented in the third
article, indicate that multispectral images captured by a drone closely
match field data and allow for effective detection of physiological
stress in trees. Surprisingly, in addition to the red, red-edge, and near-
infrared bands, the blue band also proved to be a critical indicator of
tree stress, with its effectiveness varying depending on factors such
as tree species, classification detail, and atmospheric conditions. Fur-
thermore, the values averaged over three years per sample tree, along
with the 5th and 25th percentiles of the data distribution, were found
to be particularly important because they provide a more compre-
hensive understanding of tree stress patterns over time. The use of
percentiles helps to capture the variability and extremes in the dataset,
highlighting early signs of stress that may not be visible in average
values alone. The species-specific models were then trained based on
the spectral indices, resulting in good classification accuracies (Macro
F1-Score between 0.492 and 0.769).

In essence, this thesis examines the integration of traditional moni-
toring methods with UAV-based remote sensing to enhance the effi-
ciency and effectiveness of forest health assessments. Through case
studies and empirical data, the research demonstrates how drones
can identify stress responses in trees and provide insights into forest
dynamics.

The findings suggest that drone technology offers a significant ad-
vancement in forest health monitoring, supporting the development
of targeted conservation strategies and sustainable forest management
practices. Compared to traditional methods, UAVs enable more ob-
jective assessments through high-resolution, wall-to-wall mapping,
providing full coverage of large areas at lower costs and with greater
efficiency. This approach has the potential to ensure that forests con-
tinue to thrive, provide essential ecosystem services, and contribute to
long-term economic sustainability in the face of increasing environ-
mental challenges

Object Details

• Location: Deutsche Nationalbibliothek Frankfurt am Main
• Extent: Online-Ressource
• Language: Englisch
• Notes: Universität Freiburg, Dissertation, 2025

Classification and Topics

• Keyword: Fernerkundung ; Drohne

Contributors, Places and Time

• Event: Veröffentlichung
• (where): Freiburg
• (who): Universität
• (when): 2025
• Creator: Ecke, Simon
• Contributor: Seifert, Thomas ; Tiede, Dirk ; Kattenborn, Teja ; Albert-Ludwigs-Universität Freiburg ; Albert-Ludwigs-Universität Freiburg. Fakultät für Umwelt und Natürliche Ressourcen

Further information

• DOI: 10.6094/UNIFR/262641
• URN: urn:nbn:de:bsz:25-freidok-2626416
• Rights: Open Access; Der Zugriff auf das Objekt ist unbeschränkt möglich.
• Last update: 15.08.2025, 7:26 AM CEST