Forests are complex ecosystems, and their physical, biological, and functional components interact with each other (Hansen et al., 2021), which is fundamental to maintaining ecosystem resilience and the capacity to provide ecosystem services (Watson et al., 2018). However, anthropogenic drivers such as land-use change threaten forests and exacerbate climate change and biodiversity loss (Díaz et al., 2019). Additionally, biodiversity loss may not be effectively reverted, focusing only on forest cover loss without considering other components of ecosystem functioning. For instance, fragmentation may lead to reduced habitat for animal species and significant degradation of the forest ecosystem (Morris, 2010), resulting in reduced functioning of the systems on which humans and other organisms depend (Grantham et al., 2020). Therefore, projects related to climate mitigation and biodiversity loss focus not only on one specific ecosystem service but also include the components of ecosystem integrity to deliver long-term benefits for people and the environment.

The protection of forests and the assessment and monitoring of ecosystem integrity across scales are some of the key targets of international frameworks such as the Post-2020 Global Biodiversity Framework and the United Nations 2030 Agenda (CBD/SBSTTA/24/3/Add.2, 2021). Ecosystem integrity has been defined as “a measure of ecosystem structure, function and composition relative to the reference state of these components being predominantly determined by the extant climatic–geophysical environment” (Hansen et al., 2021). It describes how complete, healthy, and resilient an ecosystem is to both natural and human perturbations (Seddon et al., 2021). The structure of the forest encompasses the three-dimensional architecture of individual plants and the connection of their attributes (Hansen et al., 2021), while the function characterizes the movement or storage of energy or matter within an ecosystem (Bellwood et al., 2019). The composition describes how a forest’s natural features are distributed within the ecosystem (genetic diversity, species richness, and community assemblages).

In recent decades the number of open-source satellite image collections has increased tremendously, which resulted in a growing number of high-quality biological remote-sensing products (de Paula et al., 2019). One well-known example is the Global Forest Cover Change map by Hansen et al. (2013). These remote sensing products are also known as Earth Observations (EO) and are particularly useful due to their global coverage and high temporal resolution (Skidmore and Pettorelli, 2015). The potential of EOs to measure and monitor biological products globally has not gone unnoticed. In 2013, the Group on Earth Observations Biodiversity Observation Network (GEO BON) set up a new framework to develop Essential Biodiversity Variables (EBVs) (Pereira et al., 2013). EBVs can be a combination of in-situ and remotely sensed data, or they can be derived from either (Giuliani et al., 2017; Schmeller et al., 2017; Kissling et al., 2018). There are currently 20 EBVs in six classes, including genetic composition, species populations, species traits, community composition, ecosystem function and structure (Hansen et al., 2021). Since the introduction of this concept, global political frameworks have proposed EBVs as the basis for monitoring advancements towards biological targets (Geijzendorffer et al., 2016). However, researchers have emphasized the importance of continuous testing of the application of EBVs across different scales and ecosystems (Pereira et al., 2013).

Reliable and consistent monitoring of forest ecosystem integrity is crucial to mitigating climate change and biodiversity loss (Keenan et al., 2015). Integrating different EBVs to monitor ecosystems globally may enable the development of a more consistent, accurate and scalable framework for sustainable management and global collaboration (Reddy, 2021). Although scientists have attempted to quantify ecosystem integrity, few studies have explored how EBVs can be organized in an effective workflow to create a composite index describing forest ecosystem integrity and differentiate between plots at the local and global scale (Hansen et al., 2021). Here, we build on previous studies defining forest ecosystem integrity by using available EBVs to present and test a framework which assesses forest ecosystem integrity of plots at the global scale. Using readily available EBVs to assess and monitor ecosystem integrity will help scientists and policymakers to acquire comparable information more easily on the state of ecosystems. Ultimately, knowing where ecosystem integrity is high or low may also help land managers and conservationists prioritize areas of high importance.

Model results from our plots (n = 333) revealed that the EII_score strongly related to the categories of forest condition (Down: n = 167, Top: n = 166). Furthermore, the EII_score was not sensitive to weight but was sensitive to plot size. Forest condition metrics, including biomass and canopy height, were correlated with the EII_score. Forests with high ecosystem integrity are located particularly in the Boreal forests and in the tropical rain forests in South America, Africa and Asia (Figure 2).

The EII_score provides an indication of how forest ecosystems across the world perform, considering ecological integrity. Our index is built on the efforts to operationalize the concept of ecosystem integrity across scales using satellite-based EOs and essential EBVs (Hansen et al., 2021). However, our approach differs from previous studies as we not only produced a global map of EII_score, but we also developed a scalable workflow that estimates EII_score at local scales but can easily be applied at larger scales. Our index showed an important property: it is insensitive to the weight used for the indicators of ecosystem integrity (structure, function, and composition). Overall, our results suggest that our index provides a reliable picture of the plots’ performance that is not driven by the importance (weight) assigned to the EII_score indicators. However, it can be used to identify the importance of area-based conservation efforts. Both characteristics are important in the context of ongoing efforts to support the monitoring and reporting of progress within the Post-2020 Global Biodiversity Framework.

The EII_score differentiated pristine from degraded forests globally, indicating that the combination of EBVs used in this study can capture different aspects of ecosystem integrity. Previous studies have emphasized the need to reconnect forested habitats to protect threatened species (e.g., Newmark et al., 2017) and thus highlight the need for an index that can measure and monitor multiple interdependent ecosystem characteristics. This study focused on differences in ecosystem integrity among plots on the global scale. Based on such an assessment, policymakers and institutions facilitating monetary incentives to conserve forested lands can make key policy decisions related to the conservation of ecosystems. However, we recognize that measuring the EII_score at other scales (e.g., global, biome, political region, ecoregion) is fundamental to halting forest integrity reduction (Keenan et al., 2015). For example, local forest management strategies will need to calculate the EII_score over much smaller scales to facilitate comparisons across forests with similar characteristics. Our EII_score has a resolution of 1 km², which is why we recommend using the EII_score only over large, forested areas to increase the variety of pixels. Additionally, this coarse resolution has resulted in some regions being assigned forest pixels even though no forest exists. This results from edge effects, which occur on the edges of plots. Edge effects are a result of pixels being square and the size of their resolution. Pixels can not be split or cut off, so sometimes the edge of a pixel falls inside or outside a plot of interest. In this example, some areas have forests, but on the edge of the forests, there is some overlap of forest pixels into areas without forests. When more detailed EBVs regarding biodiversity become available, we will be able to improve the quality of the resolution and thus reduce the edge effects.

Weights may have an impact on the value of a composite index, as weighting is strongly related to how the information conveyed by the different dimensions is aggregated into a composite index. Here we used only one indicator to represent each of the components of ecosystem integrity and considered our main underlying objective that each indicator has the same importance to ecosystem integrity. Indeed, we did not find a significant difference between the median EII_score using equal weight compared to the six possibilities of assigning different weights to the EII_score indicators. It is likely that we could have found a different result if we had used more indicators that are interrelated or if we had used more possibilities of weight distribution among the indicators than the six we used here. Note that when one indicator has weight 0, the other two indicators have weight ½, and when one indicator has weight ½, the other two indicators have weight ¼. Additionally, we tested for linearity of the EII_score with plot size (Supplementary Methods). Finally, we acknowledge that the EII_score may correlate with a broad range of metrics related to forest integrity and anthropogenic pressures. Examples of these metrics include measures related to forest condition (e.g., canopy height, biomass, structural complexity), forest ecosystem state (e.g., species diversity and abundance), and the intensity of anthropogenic pressures (e.g., land conversion). However, independently of the method used to assign a weight, weighting implies a ‘subjective’ evaluation. Like the choice for weight, we also chose to sum the pixel values to get a plot’s EII_score. As expected, we thus also found a linear relationship between the EII_score and plot size (Supplementary Figure 2). Therefore, we encourage future studies to use other indicators to assess ecosystem integrity and use different weighting distributions among those indicators. This will improve our knowledge about casual relationships among ecosystem integrity indicators and their application in different contexts.

We found that the EII_score was consistent with positive forest condition trends, including biomass and canopy height. This result thus supports the idea that our EII_score can distinguish between a healthy, thriving forest and a degraded forest (Shapiro et al., 2021). Forest degradation is often a product of human modification through, for example, land-use change, leading to forest fragmentation and resulting in reduced functioning with biodiversity loss and decreased ecosystem services (Potapov et al., 2012; Chaplin-Kramer et al., 2015; Haddad et al., 2015; Betts et al., 2019). Monitoring forests using EOs will help scientists and policymakers to identify degradation patterns and act upon them to halt or even reverse the trend. Although the results are consistent, the confidence intervals in plots with high biomass are still large making the EII_score less trustworthy in such areas. The high variation may be because the indicators we chose are not necessarily the best ones to capture this specific dimension of the ecosystem. Previous literature has found that old-growth forests could generate relatively lower values of NPP (Wang et al., 2011), while these forests have been used as indicators of high ecological integrity (DellaSala et al., 2022). Biomass, like NPP, is related to forest age (Wang et al., 2011), so this could have influenced our results.

As originally calculated (Scholes and Biggs, 2005) for one specific region (Africa), the BII provided confidence intervals consistently within 10% of the reported best estimate. Subsequently, Hui et al. (2008) conducted a more detailed analysis to disaggregate the uncertainty across taxonomic groups and biomes and found similar overall uncertainty but were able to identify mammals as the taxonomic group with the highest uncertainty as well as degraded areas and savannas.

Globally, large climatic and water availability gradients result in two orders of magnitude variation in field-measured NPP for any vegetation type on an annual scale (Running et al., 2004; Running and Zhao, 2019). Interannual variability in vegetation response to precipitation and temperature variation is estimated at 20–30% (Running et al., 2004). Validation studies show that MODIS data can largely duplicate field observations and capture observed variability in field data. Unquantifiable sources of error in MODIS data include the effects of poor weather station coverage (Zhao et al., 2006), extreme weather events, and cloud contamination, which has been estimated to differ across ecoregions.

A future next step in the development of the approach presented here is to include optimization algorithms to classify areas of high integrity within ecoregions. Our approach was based on the overlap across different global layers and EBVs to extract averaged values per pixel. Using optimization algorithms would make it possible to directly maximize the search for high-integrity pixels and simultaneously other ecosystem services or species richness. Further work should consider how businesses can use the EII_score to account for the risks and impacts of their operations on biodiversity and ecosystem services. These may include assessing the risks across a business portfolio, considering the implementation of certain projects, or producing a business counterfactual for EII_score so that the company can compare its nature and biodiversity impact against a standard baseline. From conservation and management perspectives, the EII_score can be a valuable metric to assess how different project interventions can deliver the best results considering nature conservation and social benefits for the local populations directly related to those projects.

Finally, in our study, we have used EBVs which have global coverage, but it is important to consider that there are still global biases in the availability of EBVs (Peterson and Soberón, 2018). Therefore, it is important that future studies using EBVs to account for ecosystem integrity consider carefully spatial and temporal resolutions of EBVs in order to continue improving their use to support efforts to monitor nature state such as the recently agreed Kuming-Montreal Global Biodiversity Framework. This also brings the opportunity for collaboration among countries and initiatives such as the GEOBON (The Group on Earth Observations Biodiversity Observatory Network) and fosters data availability and training capacity necessary at the global scale.

This study relied on integrating global EBVs to develop and test a framework to assess and monitor forest ecosystem integrity and health from local to global scales. Data availability, scalability, and functionality will be essential in the new Post-2020 Global Biodiversity Framework context. Therefore, the proposed methodology and EII are easily implementable and can be applied across multiple scales. The EIIscore can be used as a valuable metric for countries and businesses to quantify the impact of their actions on biodiversity and forest health monitoring. Still, further research is needed to improve methodology limitations and understand underlying dataset uncertainties. We expect that our study adds to the ongoing efforts to provide a solid ground for decision-making questions impacting the climate and biodiversity in the context of the recently agreed Kunming-Montreal Global Biodiversity Framework.

Publicly available datasets were analyzed in this study. The data and the code used in this study can be found at: https://github.com/single-earth/ecosystem-integrity-index.

AD and KV designed the study, analyzed the data, and wrote the article. KM produced the dataset and contributed to the writing of the article. M-LB, LB, KG, DK, ML, and KV contributed to the writing of the article.