Machine learning identification of microhabitat features associated with occupancy of artificial nestboxes by hazel dormice (Muscardinus avellanarius) in a UK woodland site

Hazel dormice (Muscardinus avellanarius) have severely declined since 2000 leading to increased legislative protection in the UK and Europe. Artificial nestboxes are widely used for its conservation and monitoring. Previous research has focused on how to identify suitable areas for nestboxes, but where to place individual boxes to promote occupancy is less well understood. Here, we demonstrate the use of machine learning Random Forest regression to predict nestbox occupancy from a wide range of microhabitat variables using a UK woodland as a case study. Random forest models are powerful predictive tools that allow simultaneous testing of many predictors with relatively few observations.

Field data included observed nestbox occupancy (2017-2021) and measurements of 76 microhabitat variables collected in the summer of 2021 from 45 occupied and unused nestboxes located in a deciduous woodland in Berkshire, UK. We applied Random Forest regression to identify important variables and predict nestbox occupancy demonstrating robust approaches to tune model hyperparameters and evaluate importance metrics.

In our study area, nestboxes were more likely to be occupied in sites with more hazel (Corylus avellana), greater overall tree abundance but not fully closed canopies (optimal 80-85%), more honeysuckle (Lolium periclymenum) and hawthorn (Crataegus monogyna), and when located further from footpaths and woodland margins. Occupancy over the study period was well predicted using microhabitat variables (13.3% OOB error) but future occupancy was more uncertain (33.3% error for 2021-2023 records).

Modelling approaches that allow consideration of numerous variables from few locations or observations can be help identify relevant features and predict desirable outcomes of conservation actions. Here we demonstrate this approach identifying microhabitat variables that influence artificial nestbox occupancy by hazel dormice in a UK woodland. Findings offer some recommendations for local management that could promote nestbox occupancy and improve monitoring and conservation efforts.

Methods

Study area and dormice surveys

The study area was within the grounds of Basildon Park House (BPH), a National Trust property with extensive woodland, located in Berkshire, UK (see paper for further details). 

In 2013, 144 nestboxes were erected at two different woodland areas within BPH: 78 were located in clusters within the northern site, and 66 were positioned along three distinct lines in the southern site. Within the northern site, mature beech (Fagus sylvatica), ash (Fraxinus excelsior) and oak (Quercus spp.) are the most frequent tree species with sweet chestnut (Castanea sativa) and field elm (Ulmus minor) occasional. Hawthorn and coppiced hazel are dominant in the understorey layer, with holly (Ilex aquifoliaceae) and young sycamore (Acer pseudoplatanus) frequent, and other species such as field maple (Acer campestre), whitebeam (Sorbus aria) and rowan (Sorbus aucuparia) occasional. Bramble (Rubus fruticosus agg.) and honeysuckle are abundant in both the understorey and field layer. In the southern site, nestboxes in the densest lines are closest to the boundary between woodland and arable land (in 2021 the crop was rapeseed Brassica napus), amongst what used to be a hedge, but is now a line of mature scrub. The main tree canopy species here are oak, lime (Tilia x europaea) and beech, which are frequent, although there are some sections where there are no canopy tree species present. In the understorey cherry plum (Prunus cerasifera) is dominant, with occasional European spindle (Euonymus europaeus) and young oaks. The other two, more sparse lines of boxes within the southern site extend northwards into more mature woodland, where the dominant tree canopy species are beech and oak, with hazel dominant in the understorey, and hawthorn and young sycamore frequent. Bramble is dominant in the understorey and field layer species.

All nestboxes at BPH have been regularly monitored by the Berkshire Mammal Group since 2017, up to four times a year as part of the National Dormouse Monitoring Programme (NDMP). No monitoring occurred between 2013 and 2016. During each survey, nestboxes are recorded as occupied by hazel dormice if a dormouse is present, or if there is a nest with evidence of recent occupation. Otherwise, they are recorded as empty. Using the full survey records between 2018 and July 2021 we classified nestboxes as historically occupied (occupied at least once since 2018) or unoccupied (not occupied since 2018). We did not include existing occupancy records from 2017 to reduce the time lag from recorded occupancy to habitat recorded; such that we focus on nestboxes occupied in the four years prior to recording habitat. We focused on 13 occupied nestboxes at the northern site and all 6 occupied nestboxes at the southern site (total 19 occupied nestboxes). In the northern site we focused on nestboxes occupied frequently (in more than one survey) and/or recently (occupied at least once in the last three years). If two nestboxes in close proximity (<10m) met these criteria, we randomly selected one of them to avoid replicating microhabitat data. We also selected 26 historically unoccupied nestboxes (12 in the northern site and 14 in the southern site) using the SelectRandomByPercent function in ARCGIS 10.5.1 excluding any nestboxes within 10m of selected occupied nestboxes. If two selected unoccupied nestboxes were in close proximity (<10m) we located an alternative pair in the area at least 10m apart.

Microhabitat surveys

In March 2021, all selected nestboxes were cleaned and we recorded GPS coordinates (using Handy GPS on iPhone 11, accurate to 3m), tree species on which the nestbox was installed, height from the base of the nestbox to the ground, and the orientation of the front of the nestbox. Microhabitat data were collected during May and June 2021 at four scales: directly above the nestbox, within a 5 m radius of each box, in four 2 m x 5 m quadrats starting 5 m from the nestbox, and using existing GIS layers (see paper for further details and figures). Within the 5 m radius cover was visually estimated within four levels: tree canopy, understorey, field layer and ground layer (Eden 2009). Tree canopy reflected trees taller than 4 m, with trees < 4m classified as part of understorey (Berg & Berg 1998). The four 2 m x 5 m quadrats started at the edge of the 5m radius running with orientations N, E, S and W. Sampling areas which intersected footpaths or trackways were still assessed as dormice can occasionally cross open ground when foraging or looking for nesting materials (Mortelliti et al. 2013). We acknowledge that the collected microhabitat data represent a particular time of the year and cannot capture seasonal or interannual variability, which may be important for site selection in hazel dormice.

GIS information

Nestbox locations were collected via GPS with a minimum of 3m accuracy and mapped using ARCGIS. Footpaths around the site, and the woodland margin were walked and recorded using GPS and added as a new layer. These layers were then used to calculate the minimum distance to the woodland margin, the closest footpath, and the nearest neighbouring nestbox.