Advanced Monitoring Techniques and Modelling for Tree Growth under Influence of Climate Change

Changing climatic circumstances bring more frequent and intense extreme weather events that significantly impact forests in various ways. Since forests are the largest terrestrial sinks for carbon, and are among the richest biological environments on Earth, the goals of understanding the related challenges and improving the forest resilience is high on the agenda to mitigate climate change and save biodiversity. Achieving these goals requires access to data to derive vitality and health of trees, monitor and forecast tree growth, environmental conditions data, as well as suitable data modelling approaches.

Within our research, we exploited a wide set of data sources originating and ranging from remote sensing to in-situ measurement equipment allowing us to address the tree growth and health from different spatial and temporal points of view.

The data from dendrometers provided us with the high frequency (hourly), intraday variation of tree radial growth for assessing long-term growth and instantaneous changes in growth. These data are of extreme value, as no other means to monitor trees on such a high temporal resolution with a very high sensitivity exits. However, to understand the variations in these data, which directly show variation in the tree growth, especially in the context of extreme or sudden changes, they are evaluated within the environmental context. The environmental high-quality data were collected directly from forest sites selected from the Europe-wide Forest monitoring program (ICP-Forests), which has been providing high-quality data on the vitality and adaptability of trees, nutrient cycles, water balance, etc.  

Furthermore, satellite Earth Observation (EO) data for single-tree detection and monitoring forest disturbances like selective logging and drought impacts have been likewise exploited, to explore if they may have an impact on the individual tree growth. We show that a CNN-based U-Net model trained on Very High Resolution (VHR) imagery demonstrates strong potential for identifying tree crowns and validating changes in forest structure. However, challenges such as limited training data diversity and low resolution for small trees underscore the need for further refinements.

Finally, terrestrial laser scanning (TLS) technique delivers single tree point-clouds not only allowing extraction of traditional tree features like diameters at different heights, tree height and crown dimensions, but also providing the possibility of statistical approaches for calculation of various metrics, e.g., point-cloud percentiles along the tree height and tree competition.

We describe the approaches on leveraging these, the challenges we have encountered (e.g., data gaps, errors in data, co-location), how we approached them,  and all in the context of developing predictive AI-based, climate sensitive tree growth models, to support forest management on a local, regional and national level, and thus empowering response to minimize potentially harmful consequences for modern societies in line with the UN Sustainable Development Goals.