Utilising AI to optimise stacked ecosystem service benefits in England’s public forests

Looking at flood risk reduction, environments for recreation, and habitats for biodiversity

This research aims to apply artificial intelligence (AI) to solve the challenge of how best to manage English forests for their ecosystem service provision, in particular for flood risk reduction in catchments, environments for recreation, and habitats for biodiversity. The research focuses on evaluation of the benefits and trade-offs associated with growing broad-leaved trees versus needle-leaved (coniferous) trees.

1: The application of artificial intelligence to temperate forestry and its potential for managing forests in the UK

To understand how British forestry could benefit from AI, I conducted a literature review of recent publications concerning the use of AI in temperate forest management and identified six categories of AI application: classification, inventory, growth, damage, risk, and optimisation. Classification involves the identification or estimation of a property, usually that of an individual tree. Inventory signifies a set of metrics characterising the ensemble of trees in a stand. Growth corresponds to inventory but at some stage in the future, illustrating potential growth. Damage is the detection of injury or poor health in individual trees or a stand. Risk is a prediction of where future injury or illness is likely to occur. Optimisation is the process of selecting the most efficient way to meet a forest management goal.

In determining which AI applications would be most beneficial for the specific case of woodlands in Britain, I discerned four promising uses: carbon estimation, adaptation to climate change, pest and disease monitoring, and optimising the parallel goals of timber production and ecosystem service provision. 

2: Modelling the potential ability of conifer versus broadleaf tree cover using AI in English public forests on reducing the likelihood of flooding in a catchment

Forest cover can decrease the likelihood of flooding in a catchment and previous work has indicated that coniferous and broadleaved trees differ in their effect on stream flow. To estimate the probable change in downstream flow if public forest parcels were stocked with more conifers or more broadleaves, I used SWAT hydrological modelling software. The study sites were six catchments in England with a high proportion of public forest land and with a daily record of stream flow from 2012 through 2019. Model outputs were evaluated by the difference in flow over the entire time series, by season, and during peak flows. In most catchments, the all-broadleaf scenario produced higher flow in winter and lower flow in summer, whilst the inverse was true for the conifer scenarios. 

3: Assessing the drivers of visitation in England’s public forests

This aspect of the work aimed to develop a model to 1) test whether woodland traits, particularly broadleaf/conifer mix, relate to the number of recreational users, and 2) to use real data collected on visitation rates to public forests before and during the Coronavirus pandemic to test the different assumptions in the model and determine the type of forests more attractive for visitation. 

Visitation data was drawn from records at 24 public forest car parks across England between 2015 and 2022. Predictor variables I used included trends in weather plus static properties such as population, topography, infrastructure, canopy cover, tree age, and species mix. In pairwise correlations with visitation, significant explanatory power was shown by population, rain days, mean elevation, and built amenities. I regressed these variables and percentage of broadleaf cover against visitation in a generalised linear model. The GLM had a moderately good fit but broadleaf cover was not significant. 

4: Estimating the effect of broadleaf vs conifer coverage on habitats for biodiversity in public forests

I plan to assess the potential change in bird species richness as public woodlands are converted to broadleaf or conifer. I have acquired 680,000 geolocated observations of birds in Great Britain and linked them to land cover type. Birds observed in forest were assigned a forest patch size and leaf type. I intend to plot separate species abundance curves for conifers and broadleaves, with each patch’s size versus its number of unique species. For any public forest patch of a given size, conversion to the opposite leaf type will move its species estimate from one curve to the other. 

5: Developing an AI model to optimise flood mitigation, recreational use, and biodiversity in public forests

A number of multi-objective search and optimisation methods, underpinned by machine learning, have been applied to the assessment of ecosystem services (ES). I will adapt the methodology set out by Sacchelli and Bernetti (2019) who sought to identify the thinning rotation period yielding an ideal balance between timber volume, carbon sequestration, biodiversity richness, trunk diameter, and economic gain in an Italian fir stand. My model will propose a percentage of broadleaf cover that maximises peak stream flow reduction, recreational attractiveness, and biodiversity (bird species richness) in every sub-compartment of England’s public forests.