Enhancing nature's contributions to people

Introduction

The well-being and indeed survival of humans depend on the functions of the biosphere and the benefits it provides, i.e., ecosystem services.¹ The biosphere provides a diverse array of eco- system services, including regulatory (for exam- ple, temperature regulation, clean drinking water, or pollination), provisioning (for example, food and fiber), cultural (for example, recreational, aesthetic, and spiritual values), and supporting (for example, photosynthesis, carbon sequestra- tion, or soil renewal).²

The newer concept of nature’s contributions to people (NCP) emphasizes the role of different cultures and local knowledge sources (such as local indigenous knowledge) for influencing the perception of ecosystems and contributing to shape the provisioning of services, or NCP.^3,4 However, the provisioning of ecosystem services around the world has decreased since the 1970s.² A continued loss, and potential widespread ecosystem collapse, is estimated to amount to a devastating annual cost of 2.3% of global GDP in monetary value alone.⁵ Low- and lower-middle-income countries are particularly at risk, with an estimated potential 10% loss of annual national GDP by 2030.⁵

Biodiversity is fundamental for shaping the functions of an ecosystem and its capabilities of providing NCP. Biodiversity is a broad concept referring to the variability of, for example, genes, species, functions performed by groups of spe- cies (such as coral grazing or soil renewal), and interactions (such as pollination, seed dispersal, and pest control) in ecosystems.^2,6–8 However, biodiversity is decreasing on a global scale, at a rate and extent that has been referred to as the sixth mass extinction.^9,10 This decline extends beyond a loss of species and functions to also include a pervasive erosion of genetic diversity within species.^11–13

The human realm is intricately intertwined with the natural world. Biodiversity status and NCP provisioning, or losses, are part of complex, linked social-ecological-technological systems¹⁴ that also include economic dimensions, operating from the local to global scale.¹⁵ These dynamic and interconnected systems exhibit varied responses to change, yielding both social and ecological repercussions.¹⁶ For example, mining is a form of resource exploitation and a contributor to land-use change and biodiversity decline. However, the AI hardware industry, for example, relies on mining for provisioning of critical minerals such as silicon (which forms the basis of microchips) and boron (a rare earth mineral used for example in logic and memory circuits).¹⁷

Biodiversity and sustainable NCP provisioning are vital for building capacity to mitigate and adapt to social-ecological challenges (such as climate change), support human health and well-being, and support long-term economic development and Earth System stability. AI indicates promising potential to complement traditional research methods and tools for understanding and supporting NCP provisioning. However, use of AI also comes with trade-offs, risks, and uncertainties.

AI for Data on Nature’s Contributions to People

AI methods such as machine learning (ML), deep learning (DL), natural language processing, computer vision techniques,18 and combinations thereof can contribute to NCP research in a number of ways. Specifically, to assessing, identifying, monitoring, and predicting the presence, composition and health, and people’s uses and perceptions, of ecosystems.^19–21 Audio-based (i.e., bioacoustics, ecoacoustics, and soundscape) research is one area that has benefited from AI.²² A convolutional neural network (CNN) was used for analyzing soundscape data from three urban forests in China, linking soundscapes to land use types, and assessing the impact of human activities on biodiversity.23 After large language models, multimodal language models have emerged that can analyze combined types of data. In a recent study, an audio-language model demonstrated unprecedented capacities for recognizing animal group sounds (such as birds and frogs) on par with supervised learning—but with only basic, general training.²²

Main advantages of AI for NCP include the capability of interpreting vast amounts and different types of data into training and validation of AI models.²⁴ In London, AI was used to support automated analyses of big data surveys on the prevalence and characteristics of green roofs, to increase climate-resilient green roof designs. An ML algorithm was used to segment aerial imagery at single-building level, covering 1558 km2 of Greater London.²⁵ Combinations of spatial remote sensing and social data (e.g., data from surveys or personal monitors, such as fitness trackers or social media applications such as Flickr) can help with insights on connections between socio-economic status, land use, biodiversity and ecological habitat qualities, and their spatial distribution.^26–28

Diverse data and improved models can also provide a holistic view of the human pressures on ecosystems and provide comprehensive information to guide conservation priorities. This has been shown in a study on species classified as Data Deficient in the IUCN Red List, in which a global multitaxon ML classifier predicted the probability of species facing extinction. The study re-evaluated previous threat level estimates and identified various threat probabilities, with potential implications for conservation policy.²⁹ Loss of genetic diversity and extinction risk across taxa are quantified using genetic data from environmental DNA (eDNA) and wholegenome sequencing (i.e., determining an organ- ism’s entire DNA sequence).³⁰ Here, AI could make important contributions such as classify- ing the often highly dimensional genomics data using predictive biomonitoring models.³¹

AI methods can complement traditional tools such as photo traps and manned observations to identify new patterns in data. For example, genetic data analysis enables the identification of cryptic biodiversity and indicators of ecosystem health that is not identifiable by conventional photo imagery alone.³² Sampling and analyzing eDNA from water or soil further supports the detection of elusive species including those considered extinct in specific regions.³³ More frequent sampling also enhances the ability to detect early warning signals of potential ecosystem shifts.³² Furthermore, DL approaches can now directly estimate alpha (local), beta (between-community), and gamma (regional) diversity from plot data, bypassing traditional species-distribution modeling. For example, neural networks trained on vegetation plots across Australia predicted plant diversity patterns at high resolution by learning species– area relationships from climatic, geographic, and human-impact variables, offering a scalable framework for automated biodiversity mapping in data-poor regions.³⁴

Moreover, AI can support conservation efforts and empower local communities. In the Brazilian Amazon, for example, one of the world’s most biodiverse regions, Indigenous groups have used drone data and AI-based monitoring tools to protect their territory against invasions of illegal miners.³⁵ Digital AI applications based on citizen science, in which people contribute with data and identifications, such as iNaturalist and Flora Incognita, can help identifying different individual plant species and contribute to our understanding of plant trait patterns on a global scale.³⁶

AI-Assisted Predictive Modeling of Nature’s Contributions to People

Modeling of NCP and biodiversity is primarily conducted using ML, DL, and computer vision techniques. Potential outputs include predictions, pattern and trend detections, and impact assessments across spatiotemporal scales.³⁶

AI’s ability to analyze complex remote sensing data and identify key indicators increases the availability of data for modeling scenarios. AI can extract and evaluate indicators (such as changes in vegetation cover, or air and water quality) from remote sensing images and create climate adaptation scenarios that benefit both humans and ecosystems.²⁵ Such capabilities can make significant and potentially even lifesaving contributions. For example, by mitigating the impacts of natural hazards, or contributing to NCP-enhancing design suggestions for natural (green-blue) infrastructure in cities.²⁵

The predictive capabilities of AI can inform the analysis of complex data such as the combined effects of multiple interacting stressors. For instance, AI has been employed to evaluate the cumulative impact of climate and land-use change on NCP provisioning in Yunnan Province, China.³⁷

Earlier ecosystem models have demonstrated a capacity to identify thresholds (such as removing 80% of biomass in an ecosystem³⁸), for sudden and often irreversible changes with cascading effects, known as “tipping points.”³⁹ Recent advancements in AI also enables integration of species abundance, genetic diversity, and functional traits within phylogenetic frameworks, enabling multidimensional predictions that capture complex eco-evolutionary dynamics beyond traditional threshold models^40,41 (see also Theme box 2, Using AI to Detect Earth System Tipping Points).

AI methods can augment human capabilities and assist discerning patterns in complex social-ecological data.⁴² For instance, CCNs in species distribution models (SDMs), such as those relating to pollinator diversity, can extend beyond site-specific predictions to also estimate the potential influence of surrounding landscape structures. This capacity is particularly useful in the study of highly heterogeneous landscapes such as urban and peri-urban areas.⁴³ Other predictive studies have used ML to assess the suitability of plant species based on their adaptability to projected future conditions.⁴⁴

Studies on land use and land-use change have benefited from AI, in part due to ML’s ability to analyze large and high-resolution datasets, such as satellite data. For example, recent AI-supported analysis at a 1 km resolution indicates that 63% of urban expansion will likely occur on current cropland, which could decrease the global crop produc- tion by 1–4% (equivalent to the annual food requirements of 122–1389 million people).⁴⁵ Large-scale biodiversity data analyses supported by AI methods have enabled new forms of modeling of global microbial soil systems (crucial for soil renewal and thus supporting NCP). The approach has enabled projections of ecological functional and diversity responses to perturbations, such as climate change, invasive species, and pollution.⁴⁶

AI-Assisted Decision-Making for Nature’s Contributions to People

Participation in land management and decision- making can foster a sense of meaning and responsibility toward the natural environments,

i.e., stewardship —an important factor for ecology conservation and long-term support for NCP.⁴⁷ AI can support decision-making concerning NCP at various levels—for example, mobile applications that can advise small-scale farmers or citizen gardeners on everything from sowing to harvesting^48–50—while larger-scale efforts include guiding conservation policy,⁵¹ biodiversity protection, and land management.⁵² One example of AI research providing actionable advice to mitigate ecological and social harm is a study of the wetlands of Kolkata, India. The AI analysis of combined biogeophysical and social data revealed that 60% of the wetland’s NCP provisioning was threatened.⁵³ The information can support planners in assessing preferred sites for, and characteristics of urban development, and for conservation areas.

AI can inform decision-making processes by identifying nuances in social-ecological data. Examples include mapping how people use ecological resources such as gardens,⁵⁴ assessing NCP, such as the stormwater absorption capacity of green infrastructure,⁵⁵ and understanding different perceptions of ecology and NCP.^56,57

AI-driven analyses of crowdsourced and citizen data, such as photos of natural areas or fitness tracker data, can inform decision-makers about how people utilize and value natural spaces.^28,58

Potential

AI has the potential to contribute to the analysis and management of NCP in various ways, particularly in areas where research has previously struggled. Promising application areas include: a) integrating different data such as specimen location, habitat preferences, ecological data, and spatial datasets on human impact, to advance social-ecological predictions and maps,⁵⁹

1. b) evaluate ecological and environmental policy and practice by combining different types of AI, such as natural language processing and computer vision, to enhance the understanding of how natural spaces and resources are used and monitor changes⁶⁰ c) improving remote sensing analyses, e.g., for cover classification⁶¹ and threat detection,⁶² and d) accelerating analyses of images and recordings.⁶³

Exciting opportunities also lie in advancing AI-driven scenario modeling of NCP-based strategies. Specifically, for strengthening the adaptive and mitigative capacities of both ecosystems and human societies in response to, for example, climate change and biodiversity loss.

Limitations

The geographical bias identified in our literature review indicates that a known issue in ecological research⁶⁴ persists in AI applications for NCP. By predominantly focusing on upper-middle- and high-income countries, AI-generated NCP outputs, such as recommendations on ecosystem management, risk misaligning with the perspectives of people and the ecological, climatic, and socio-economic contexts in other regions, i.e., the Majority World.

Several of the less studied countries, such as Brazil, Colombia, Indonesia, and in sub-Saharan Africa, often have limited AI capacities. They are also exceptionally rich in biodiversity, with a high number of threatened species,⁶⁵ and with the highest number of people who depend directly on ecosystems for their livelihoods. For instance, it is estimated that over 50% of Africa’s workforce is employed in agriculture.⁶⁶ However, even in Africa’s burgeoning AI sector, NCP and biodiversity risk being overlooked when other areas are prioritized.⁶⁷

Predictions of NCP, such as the decomposition of organic matter, pest control, crop pollination, and seed dispersal for agriculture and habitat maintenance, require modeling of both biophy- sical variables and of biodiversity.⁶⁸ However, these two areas have largely been studied separately, and biodiversity information is often lacking in NCP predictions.⁶⁹

AI-generated insights on NCP alone do not guarantee sound decision-making.⁷⁰ Preferences and resource management histories vary across groups,^71,72 yet underrepresented groups–such as women, the elderly, Indigenous peoples, and low-income communities–often lack visibility in both data and decision-making.^73,74 Non-traditional sources like social media risk reinforcing in- equalities by reflecting the behaviors and values of user groups with access to specific digital applications. Access to, and use of digital applications vary by gender, geography, age, income level and so on, and not everyone uses digital platforms or has reliable internet access.^75,76

The dynamic nature of NCPs, as they arise from interconnections between humans and the biosphere, combined with data gaps, can lead to differences between AI predictions and actual outcomes. This phenomenon is known as “predictive dissonance”.⁷⁷ Such dissonance could undermine trust in AI NCP recommendations, which is one reason why it’s important to clearly communicate uncertainty factors in AI modeling outputs.

Only around half of the models in NCP predictions have provided quantifications of uncertainty,⁶⁸ and NCP research suffers several blind spots.⁷⁸ Analytical uncertainty in NCP modelling can occur in several steps of the research process: in data quality and biases,⁷⁸ data preprocessing and supervised classification,⁷⁹ the selection of predictor variables,⁸⁰ the behavior of model algorithms, and decisions concerning baseline data, to name a few.⁸¹