Project Highlights:

  • Use cutting-edge machine learning techniques to derive relationships between drought and tree mortality from observations.
  • Apply this new knowledge in a state-of-the-art global forest model to understand the implications for the global carbon cycle and climate
  • Work directly with contributors of forest plot data to understand its potential and limitations.



Why do trees die? How do the rates and drivers of tree mortality vary from the tropical rainforest to the boreal? How do these mortality rates affect forest structure? These questions are crucial in order to understand large-scale forest dynamics. They are also of fundamental importance for our understanding of how climate change will evolve because forest ecosystems are huge stores and sinks of carbon.

Hot droughts are expected to become more prevalent under climate change (Seneviratne et al., 2012), resulting in a lot of focus in recent years on how trees die under drought (Allen et al., 2015; Hember et al., 2016; Rowland et al., 2015). Yet little is known about the extent to which drought usually plays a role in tree mortality in the different ecosystems around the world. For instance, is drought a more prevalent driver of tree mortality in areas that are commonly hot and dry, or in those where such conditions are a relatively rare occurrence? Which types of tree are most vulnerable and how is this affected by their status in the ecosystem?

This project will apply the latest machine learning techniques developed for genomics problems (Basu et al., 2018) to identify patterns and relationships between drought and tree mortality in of a range of forest inventory and remote sensing datasets, spanning global forest biomes. This new process-knowledge will then be integrated within a state-of-the-art global ecosystem model (Smith et al., 2014), to understand the implications of drought-induced tree mortality for current and future forest structure and carbon cycling, thus achieving a complete flow from observations, to understanding, to implications.

This project has the potential to substantially advance understanding of how drought interacts with forest at a biogeographical level – the scale that is most relevant for understanding feedbacks of tree mortality on future climate change.

The studentship collaborates with Operation Wallacea, a major source of inventory data from the tropical forests, which will be analysed for the first time for tree mortality within this project. The student will be part of a larger team working on tree mortality through the TreeMort project (more.bham.ac.uk/treemort).

Figure 1: Drought-induced tree mortality has been recorded all over the world (figure from Allen et al., 2015; dots and ovals indicate location of recorded drought-related tree mortality events in recent decades), but to what extent is this something out of the ordinary vs normal background dynamics?


The project draws on big data of tree growth and mortality which has been collected through national forest inventories and scientific research plots, such as those maintained by project collaborator Operation Wallacea, and from satellite remote sensing. Machine learning methods will be applied to identify links between observed mortality and environmental drivers and a new algorithm leveraged to draw well-defined relationships from these linkages (Basu et al., 2018). These techniques are much more powerful than those applied in previous work because of the lack of prior assumptions and ability to pick out highly non-linear relationships.

The new tree mortality relationships developed will be integrated within the LPJ-GUESS global ecosystem model (Smith et al., 2014) to understand how forest dynamics are influenced by drought mortality and how this is likely to change in the future. This will link closely to on-going work to improve processes underlying drought mortality in the model.

Training and Skills

The student will participate in an Operation Wallacea field campaign (e.g. Madagascar, Sulawesi) in order to gain understanding of the realities and limitations of gathering experimental data in the field.

They will develop skills in big data handling and machine learning, along with quantitative ecosystem modelling, scientific code development and data analysis. There will be substantial opportunities for travel and to build strong international links, collaborating with partners from across Europe.

The Doctoral Researcher will be supported in developing research communication skills, including presenting their work at international conferences and publishing their findings in high-impact journals.


Year 1: Familiarisation with concepts and data. Field trip to forest plot site. Analysis of Operation Wallacea plot data (Paper 1). Training in machine learning techniques. Preparation of satellite and ancillary additional datasets for global analysis.

Year 2: Global-scale machine learning analysis of tree mortality and drought interactions (Paper 2). Presentation of results at international conference.

Year 3: Ecosystem model reparameterisation and application to understand implications for forest structure and carbon cycling (Paper 3). Presentation of results at international conference. Final writing up of thesis.

Partners and collaboration (including CASE)

The studentship is offered in collaboration with Operation Wallacea who have a unique forest inventory dataset for several tropical locations which has never been analysed for tree mortality. This offers an excellent opportunity to analyse a new dataset in close collaboration with those who have collected it, before expanding the methods to take in a wider range of datasets and forest types.

The student will benefit from the large network of international collaborators taking part in the TreeMort project, based at Birmingham. They will thus be part of a much larger team working on the broad topic of tree mortality.

Further Details

Please add project/institutional contact details including a link to the application website if applicable