IAP-24-041

Rewilding: integrating technologies and data to monitor upland food-web change

Extensive ecosystem restoration is increasingly seen as being central to conserving biodiversity [1] and stabilizing the climate of the Earth [2]. Central to restoration success is effective biomonitoring to assess outcomes. Rewilding, defined as “the reorganisation of biota and ecosystem processes to set an identified social-ecological system on a preferred trajectory, leading to the self-sustaining provision of ecosystem services with minimal ongoing management” [3], is gaining momentum as an environmental management option in many upland areas of Britain, but other competing land use demands, driven by recent changes in UK policy, has made understanding the consequences of management decisions about the future of the uplands urgent [4]. Yet the scientific evidence required to make such decisions, in particular how rewilding will affect upland biodiversity (especially species of conservation concern), is lacking. Furthermore, monitoring biodiversity outcomes (measures/metrics) over long timescales is a significant challenge, although new biomonitoring technologies are developing rapidly.

The Glen Finglas landscape-scale grazing experiment was established in 2002 to examine how changes in sheep and cattle stocking densities affects upland bird and animal communities. After 15 years of experimental grazing treatments, significant changes in plant and animal communities became evident [5], including a significant increase in bird species richness [6]. Thus, vegetation structural changes as a result of livestock removal are likely to affect other trophic levels within the food-web, but this is yet to be holistically examined. This project will test and develop novel, integrated biomonitoring methods, aided by Deep Learning (DL), to constructing highly- resolved species-interaction networks that incorporate plants, arthropods, mammals and birds. Uniquely, the project will investigate how livestock grazing treatments affect the structure, complexity and robustness (i.e., the attack tolerance of the network to species extinction) [7] of upland food-webs, providing policy-relevant evidence (and potentially new metrics) on how changes in upland management affects ecosystem resilience.

The project has three interlinked objectives:
OBJ. 1: Integrate biomonitoring technologies to construct upland food-webs consisting of plants, arthropods, mammals and birds based on observed and inferred species-interactions collected over a 20-year period.
OBJ. 2: Compare the structure, complexity and robustness of upland food-webs under different experimental grazing treatments.
OBJ. 3: Predict long-term management outcomes for the uplands using adaptive network models that incorporate long-term data.

There will be opportunities for fieldwork at Glen Finglas with trained staff to deploy biomonitoring technologies (i.e. drones, soundscapes and image-recognition), and/or test specific hypotheses devised by the student and supervisory team.

A replicated, randomized-block experiment consisting of six replicates of four treatments was initiated in 2003 at Glen Finglas, in central Scotland, with baseline data collected in 2002. Plots are each approximately 3.3 ha in size with altitudes ranging from 200-500 m above sea level. Treatments of [I] 9 ewes per plot (2.72 ewes ha-1), [II] 3 ewes per plot (0.91 ewes ha-1), [III] 2 ewes per plot (0.61 ewes ha-1) with autumn cattle grazing to give equivalent offtake to II, and [IV] ungrazed, were randomly allocated to plots within each block. Within each plot, plants, small mammals (field voles) and breeding birds are surveyed each year, with more intensive fieldwork (e.g., foliar arthropods) carried at regular periods (but not annually). Other data (e.g. foxes, nocturnal moths, carabid beetles) are available from more focussed studies that have occurred over the past 20 years.

The project will pool and process the existing plant and animal data that has been systematically collected over a 20-year period. Species-interactions will be determined based on observed interactions in the field (e.g., insect flower visitation, prey provisioned to nestlings), recorded interactions in public repositories (e.g., UKCEH Database of Insects and their Food Plants [DBIF]) and probabilistic models for species co-occurrence combined with novel DL methods developed by the team. Food-webs will be created for each plot and across years, following a similar methodology that was recently used to construct potential landscape-scale pollinator networks across Great Britain [8].

The project will test and a develop a) ways of integrating multiple biomonitoring technology sources using deep learning – specifically Large Language Model approaches that maximises the multiple sources of data to infer interactions to examine changes to biodiversity and ecosystem functioning [9], and b) a range of contemporary network analyses, especially Bayesian approaches, to examine the effects of livestock grazing pressure on network complexity, stability and robustness. The experimental design is well-suited for hierarchical modelling. The project will also evaluate the application of dynamic network models for predicting long-term management outcomes, with a particular focus on rewilding (i.e., when livestock are completely removed from upland habitats).

There are opportunities for fieldwork at Glen Finglas with staff from the James Hutton Institute (JHI), including a placement at JHI (Aberdeen) where the long-term data is curated and stored.

Year 1

The student will be introduced to the 20-year dataset and receive training in food-web construction methods, Deep Learning and ecological network modelling. Fieldwork including deployment of camera traps, AudioMoths and invertebrate biomonitoring technologies, and supplementary plant and animal sampling at Glen Finglas to fill identified ‘interactions gaps’ in the dataset.

Year 2

Validation of interactions based on species co-occurrence models. Modelling the impacts of livestock treatments on the upland food-web, focussing on network structure, complexity and robustness (contrasting qualitative and quantitative structures and using Frequentist and Bayesian approaches). Some hypothesis testing at Glen Finglas using a sub-component of the network. First publication.

Year 3

Training in adaptive network models, parameter testing and analysis using the long-term dataset. Forecast management outcomes using network predictively. Second publication. Communicate results to stakeholders, the public and academic community.

Year 3.5

Thesis writing and final publications,

The student will have full access to Newcastle’s SAgE Faculty Research Development Framework (RDF). This programme provides training in the four key domains of the Vitae RDF including: knowledge and intellectual abilities, personal effectiveness, research governance and organisation, and engagement, influence and impact.

Specifically, the student will benefit from working in a team consisting of applied ecologists (James Hutton Institute), network ecologists (Newcastle University) and ecological modellers (St. Andrews) on a novel and policy-relevant topic. The project will specifically provide skills in: 1) ecological census techniques; 2) network construction (including Deep Learning with support from the School of Computing) and bioinformatics using Newcastle’s Rocket HPC; 3) bespoke network analysis and modelling; 4) science communication (for a range of audiences, including articles in The Conversation). Further training in statistical methods is available at JHI through Biomathematics and Statistics Scotland.