StatisticsLab

The StatisticsLab is a think-tank, to foster open discussion and development on a range of subjects (see below) in the field of statistics applied to forest inventorying and monitoring. It is a scientific forum open to the contribution from all interested collaborators within the RU Forest Resources and Management. It promotes discussions on a range of topics at the interface between current knowledge and the further development and needs for statistical methods. The topics include (but are not restricted to)

  • Methodological questions related to forest inventorying and monitoring
  • Sampling theory
  • Spatial and time-series analysis
  • Estimation methods

The members of the StatisticsLab (i) meet on a regular basis to discuss questions related to the topics mentioned above, and (ii) organize annual open Workshops/Seminars for all those interested in the subject.

Contact

Regular Meetings in 2025

  • February 26th (Bi-EPD1), 13 – 15 h
  • May14th (Bi-EPD1), 13 – 15 h
  • August 20th (Bi-Flurysaal), 13 – 15 h
  • November 19th (Bi-Flurysaal), 13 – 15 h

StatisticsLab Colloquium 2025

October 28th (Bi-Englersaal), time pending

Past StatisticsLab Colloquia

Tree mortality beyond field detection: challenges for inference, modeling and predictions, October 24th, 2024

Jeanne Portier from WSL, Forest Resources and Management, presenting: Modelling tree mortality from forest inventory data: which approach for which purpose?

In recent years, increasing background tree mortality has been observed across many European forests. Numerous studies investigate how forests are responding to global changes to foresee how these forests will develop in the future, and to generate management recommendations accordingly. Forest inventory data are often used in such studies as they can provide representative data at regional to national scales. Modelling approaches based on such empirical data can help identifying and understanding the drivers of tree mortality, as well as predicting current and future forest development trajectories. However, forest inventory data are complex and their statistical handling presents some methodological challenges. In addition, tree mortality data are highly unbalanced (higher proportion of living trees compared to dead trees), which further complicates the modelling process and performance assessment. For these reasons, while the drivers of tree mortality are increasingly well understood, we are still lacking concrete and practical guidelines on how to develop well-performing empirical tree mortality models that account for the complexity of forest inventory data.
We used Swiss National Forest Inventory data (~4300 plots, up to 5 times remeasured over the last 40 years) to identify the characteristics of forest inventory data that should be accounted for when modelling tree mortality. We determined if and how these characteristics could be implemented in various modelling approaches such as generalised mixed models, a survey approach, random forest and convolutional neural networks. We compared the performance of these approaches when inferring the effect of environmental factors on tree mortality as well as when predicting tree mortality. We conclude that the best-suited modelling approach depends on the goal of the model, i.e., inferring or predicting. We stress that metrics used to evaluate the performance of a modelling approach must be adapted to unbalanced data, and provide recommendations on best practices for modelling tree mortality from forest inventory data. Our results can help future tree mortality studies to choose, based on their goal, which modelling approach would be most suitable as well as the appropriate metrics to evaluate the model’s performance.

Stefan Hunziker from WSL, Forest Dynamics, presenting: Tree mortality observations on an annual time scale and the relationship with crown defoliation

As part of the Long-term Forest Ecosystem Research (LWF), the condition of tree crowns is monitored on an annual time scale. This makes it possible to follow the defoliation trajectories of individual trees until they die and to consider a wide range of stress factors that may have played a role in this process. I will give a brief insight into research questions such as: How has tree mortality in Switzerland evolved over time? How are mortality and tree crown defoliation related? Are there tipping points and can crown defoliation serve as an early warning signal for increased mortality? What are the main drivers of increasing crown defoliation and ultimately tree mortality?

Ross Shackleton from WSL, Forest Resources and Management, presenting the documentary movie: Our Changing Forests

Since 2015, Europe’s forests have faced increasing defoliation (leaf loss) rates — which is a key sign of the declining health and vitality of trees. In this short film, we explore this with researcher Ross Shackleton in Zurich's forests, who shows us examples of different levels of tree defoliation and discusses what is causing this. Dr Marco Ferretti further explains the alarming trends of increasing defoliation rates across Europe the implications of this for our environment and society as well as options to make forests more resilient in the future. (Watch here)

 

The effect of spatial structure of forests on the precision and costs of plot-level forest resource estimation, May 30th, 2023

Mari Myllymäki from the Natural Resources Institute Finland (Luke) Helsinki

An important goal of forest inventories is reliable estimation of growing stock volume or biomass within some predetermined area, which may range from some hectares to a whole country. Although the inventory methodologies are under constant development, in particular because of increased utilisation of remote sensing information, accurate and precise field measurements are of crucial importance for statistics and also for remote sensing aided inventories, e.g., as training data. Field measurements are expensive wherefore the traditional question has been and still is how to allocate the measurements for an optimal use of resources. In this talk, I will present a study that was made to investigate how the accuracy and costs of forest resource estimates for sample plots of different size and shape depend on the spatial pattern of trees, and on tree density and size distribution. We used empirical data from 396 mapped forest plots from Finland. The variance of the unbiased Horvitz-Thompson estimator and expected costs of the basal area and tree density estimation were calculated for 99 different sample plots of different type and size in each of the mapped forest plots. I will discuss the results, and possibilities and limitations in utilizing them, e.g., in large scale national forest inventory. More information in Häbel et al. 2019