Inverse modelling of invasive forest pests

This project aims to use historic data to predict and understand future invasions by non-native forest pests. We develop new methods to predict establishment and spread locations based on limited available data.


Invasive forest pests are non-native species feeding on woody plants. These species are generally introduced via trade in woody materials or packaging. Once established in a climatically suitable area with suitable host species, these pests may spread to other areas. This spread process is mainly driven by trade in for example firewood or other anthropogenic means.

The resulting impact ranges from damage to urban trees, forests for wood production or biodiversity and permanent crops such as olive trees; and from relatively minor damage up to near extinction of tree species. Man driven factors are of such importance that non-native pests generally occur in areas with high anthropogenic activity. When analyzing data of these pests however, a problem arises. Since pests are observed by humans, they are also more likely to be observed in areas with a lot of people. This sampling process can bias the available data, and this has to be considered when creating models. 

Project description

This project aims to develop a full suite of tools for the management of novel pests. Our part in this project aims to create novel methods to analyze the limited available data on historic invasions to make predictions on future invasions.  

A major contribution is in the hotspot model for forest pest establishment in Europe. The method used to create this model compares the distribution of environmental characteristics at areas where pests first established to the distribution of these environmental characteristics over Europe. This gives us insight into which characteristics lead to a high probability of establishment, and which locations are likely. This method falls in the category of so called “species distribution models”. What is novel to this method is the flexibility in how to construct the individual model components. This allows the modeler to make the model more or less complex, include ecological knowledge and test hypotheses related to distributions of environmental variables and species traits. This method also allows explicit modelling of variable covariance: how variables relate to each other. For example, an area close to a city tends to also have a higher population density. Correcting for sampling bias is also possible when information about the sampling process is available.       

Pest dispersal at the continental scale was also studied and a novel method was developed to handle sparse data of varying quality. This is important since a major issue is that pest observations come from varying sources and are inconsistent. To construct a model, historic spread data is combined and the distances between sightings of following years calculated to fit ‘dispersal kernels’. These functions predict the probability of dispersal given the distance between source and destination areas. Dispersal kernel functions can also be used to study the influence of known drivers. For example, areas of high human population density may stimulate pest dispersal to surrounding areas.


We are currently working on three separate research articles which are expected for 2022:

  • an introduction of the new hotspot method with comparison to an existing methodology,
  • hotspot modelling of invasive forest pests,
  • estimating spread rates of invasive forest insects in europe with a general dispersal kernel.