PhD defence - Yan Cheng

Remotely Sensed Forest Health - Tree Mortality Mapping and Characterisation using Sub-meter Aerial Images and Computer Vision

Assessment Committee

Associate Professor Andreas Westergaard-Nielsen (Chairperson)
Senior lecturer Thomas Pugh
Associate professor Mariana Belgiu

Supervisors

Supervisor: Associate Professor Stephanie Horion, IGN
Co-supervisor: Professor Claus Beier, IGN

Place

The defence is conducted as a hybrid defence.

To attend the digital defence, please follow the link:

https://ucph-ku.zoom.us/j/61088413465?pwd=WXL2XAeB5MS2KBfz0KqZgtmctTmVbF.1

Instructions if you wish to attend the defence via the digital solution: Please follow the link and hereafter the instructions to download the required -client. If the -client is incompatible with your pc, smartphone etc. you can attend via an Internet browser. Log-in in due time before to allow time to install the -client.

The physical place of the defence:

Department of Geosciences and Natural Resource Management, Building: ØV10, Area 6, Ground floor, Auditorium C, Address: Øster Voldgade 10, 1350 Copenhagen K

The defence is followed by a reception in Røde Stue (Red Room), ØV10. Everybody is welcome.

Ask for a copy of the thesis here: yach@ign.ku.dk

Abstract

Climate change has led to global changes in the spatial distributions and compositions of forests (Parmesan and Yohe, 2003; Kelly and Goulden, 2008). Trees cannot relocate as individuals; these shifts in forest distributions are partially a result of excessive tree mortality in regions where environmental conditions are no longer favourable for them. In extreme cases, forests have turned into drylands or even deserts. Forests, having existed much longer than human beings and survived drastic climate transformations, can adapt an recover from disturbances but may take up to a millennium to return to the initial state (Trumbore et al., 2015). The real threat from forest disturbances lies in their impact on us. Our survival is intertwined with that of forests, relying on their resources and their roles in regulating ecosystems and the global climate, and we cannot afford to wait a millennium. Planning for resilient forests is, therefore, also about planning for our own survival as we face the escalating effects of climate change caused by ourselves.

This thesis was driven by the need for comprehensive, large-scale maps of tree mortality, which are critical for understanding tree mortality dynamics and guiding resilient and sustainable forest management (McDowell et al., 2015; Trumbore et al., 2015; Hartmann et al., 2018). It is structured around four main papers, each building upon the previous to address this gap. Paper I revealed that most dead trees in California, a region heavily impacted by various natural disturbances, are scattered or in small clusters, which highlights the necessity for high-resolution tree mortality mapping. This insight was derived from maps of individual dead trees generated by applying Artificial Intelligence (AI) models to sub-meter resolution airplane images from national aerial surveys. Paper II extended beyond a single region to develop generalizable AI models that can handle image variations across countries, paving the way for continental-to-global mosaics of high-resolution tree mortality maps. Paper III further broadened the focus beyond completely dead trees to include the detection of partial canopy diebacks using higher-resolution drone images, enabling the early detection of tree stress signals and allowing early human interventions to prevent irreversible damage. Paper IV addressed the scarcity of training labels in AI-based Remote Sensing (RS) applications in forestry, by collecting diverse high-resolution aerial images from most tree mortality sites worldwide. Combined with an interactive platform and the integration of pre-trained tree mortality detection models, this initiative fosters collaborations among plant scientists, foresters, RS scientists, and computer scientists, contributing to an improved understanding of the global dynamics of tree mortality. Additionally, a co-authored paper in the supplementary illustrates how high-resolution tree mortality maps support understanding the drivers of tree mortality in urban areas, showcasing the integration of such datasets with others can potentially yield more detailed ecological insights.

In conclusion, this thesis demonstrates the necessity and feasibility of high-resolution tree mortality maps at large scales, contributing to this field by advancing technical aspects and fostering interdisciplinary and international collaborations. Mapping is a crucial starting point towards a global understanding of forest health and resilience. Without a comprehensive knowledge of where tree mortality occurs, further understanding and actions may be limited. High-resolution maps also enable more comprehensive follow-up studies, which can improve our knowledge of the processes and impacts of tree mortality. This, in turn, can guide better decision-making and policymaking in forest management and recovery strategies. Ultimately, this work contributes to guiding the building of more resilient and sustainable forests, which matters to the well-being of humankind and other species relying on these ecosystems.