Measuring, Reporting, and Verification of Forest Restoration

2.4.1 Rationale
There is a large body of scientific literature assessing the habitat requirements of forest birds, and the impact of production forest management on community composition and species populations. Very few bird species in Europe are dependent on a particular tree species. Some notable exceptions include the preference of Black Grouse (Tetrao tetrix) for birch buds, a particular race of the Nutcracker (Nucifraga c.caryocatactes) depends on hazel nuts, and Brambling (Fringilla montifringilla) is highly dependent on beech seeds in winter. Nevertheless, bird species, often exhibit a preference for particular categories of trees (i.e. coniferous versus broadleaved), or particular compositions of tree species at appropriate spatial scales. The long-tailed tit (Aegithalos caudatus) is more commonly found in landscapes with over 15% deciduous cover, whereas Coal Tit (Parus ater) and Goldcrest (Regulus regulus) are common in forests with at least 10% spruce. Although there is little doubt that greater bird diversity is associated with greater tree diversity (Ampoorter et al 2019), and therefore compositional shifts in birds communities could be anticipated to result from many restoration projects, not all species will benefit and there are many forest characteristics within the range of management control that will influence the species that colonise new habitat (Penone et al. 2018).

Bats are often forgotten by members of the public, but arguably play as important a role in maintaining forest ecosystem resilience, including control of insect pest within forests and beyond. Like birds, bats forage at large-spatial scales, and their presence depends as much on the availability of habitat and connectivity at the landscape-scale as it does on individual stands. The richness of European bat species increases with broadleaved forest cover in temperate and boreal regions (e.g. Meramo et al. 2025), and with conifer abundance in Mediterranean landscapes, these changes in vegetation structure in turn reflecting the underlying changes in insect abundance (Barbaro et al. 2019). Often the conservation of rare birds and bats are a conservation objective in their own right, but more generally they are valuable additions to a monitoring plan because they have differing habitat preferences and requirements, particular for nesting and roosting, that provides complimentary information about forest restoration success.

2.4.2 Survey methods
Although audio has been widely used to study birds and bats for many decades, in the last couple of decades cheaper digital devices have enabled passive acoustic monitoring (PAM) to flourish as an accepted technique to monitor many sites over prolonged periods of time (Gibb et al. 2018; Sugai et al. 2018). PAM is suited to automated data collection and processing, so large amounts of data can be readily analysed, and the factors affecting detection easier to quantify than for human observers (which depends on expertise) and therefore especially useful when repeated surveys may span many years. While continuous monitoring is unlikely to be justifiable for most restoration projects, the prospect of networked sensors and on-board analysis pipelines raise the possibility of real-time PAM monitoring and adaptive management in the future, that can further act as input to public engagement (Roe et al. 2021; Sethi et al. 2018).

PAM enables systematic long-term comparable data collection, as long as detailed protocols of recording schedules are attached (Gibb et al. 2018). PAM recorders are typically limited by their power and storage capacity, and sampling effort in acoustic monitoring can therefore optimise the distribution of acoustic sensors and recording schedules to maximise monitoring periods and decrease maintenance requirements (Sugai et al. 2019). As described in section 1.4, if a reference model does not already exist to guide monitoring design, pilot studies can rapidly determine the number of recorders and the duration of surveys best suited to the ecosystem (Metcalf et al. 2023; Sugai et al. 2019).

2.4.3 Data collection to data reporting
Acoustic monitoring generates large volumes of data, (e.g., GB per day per recorder) that necessitate automated processing and analysis. One approach to minimising computation is to consider the soundscape as a whole (Bradfer-Lawrence et al. 2023. A focus on broader soundscape patterns may be preferable when restoration objectives are focused on the integrity of habitat quality and ecosystem functioning, rather than specific species (Bullock et al. 2022). Nevertheless, while there is broad agreement soundscape change is consistently a good indicator of community change, it has proven difficult to identify consistent soundscape changes (e.g. Sethi et al. 2023), and they are regarded as less tangible objectives for land managers and funders alike (Cord et al. 2025).

Another less intensive approach to analysis that does not require significant expertise is to allow programs to group all similar sounds, and compare how the diversity and frequency of those types change. Manual review is typically required to rule out the possibility sounds are derived from anthropogenic activity. Importantly, unique sound types could include calls from multiple species, or multiple distinct sounds could all be from the same species. Although easily scalable, this method has not been widely applied, likely because results are not easily comparable among studies, and because stakeholders are typically interested in the identity of specific species.

Finally, the most detailed approach is to apply automatic species detection and classification software. For very large datasets automated classifiers are the only viable means of generating species-level information that can be equated to traditional metrics like species richness and community composition. Machine learning classifiers do still make mistakes as many species can sound similar, or modulate their calls depending on behaviour, but increasingly advanced machine learning methods are constantly improving the state of the art (Kahl et al 2021; Google Research 2023; Huus et al. 2025). For example, BirdNET is trained to identify over 6000 bird species (Kahl et al 2021) and has been shown to effectively identify their presence in large datasets (Funosas et al. 2024), particularly for cryptic species (Bota et al., 2023). Note that the software identifies sounds, but without designing an acoustic system specifically to detect the direction a sound came from, the frequency of acoustic detection does not offer a good proxy for abundance or density. Furthermore, while large machine learning models are optimised to minimise the amount of post-processing verification required (Pérez-Granados, 2023), the main issue remains how to minimise the influence of errors that can substantially skew assessments of community composition and richness. Thus, although PAM is undoubtedly a powerful technique for surveying birds and bats that helps standardise monitoring across space and time and offers a practical and scalable solution that balances cost and ecological insight, users need to remain wary of reporting outcomes that are entirely automated. As a result, current applications of PAM are typically semi-automated, involving regular manual cross-checking to resolve ambiguous classifications within analyses (Campos-Cerqueira & Aide, 2016; Doser et al., 2021). By doing so, the presence of key species, such as those of conservation concern, can be efficiently verified (e.g., Wimmer et al. 2013).