Introduction
The ecosystem (bolded terms defined in S1, Supplementary Information) is a major focus in several prominent conservation strategies established to curb impacts to biodiversity. For example, the Kunming-Montreal Global Biodiversity Framework (GBF) (CBD 2022a) identifies the maintenance or restoration of ecosystems as a long-term goal. The GBF, and national strategies issued by GBF signatories (e.g., ECCC 2024), includes targets for realizing this goal, and those targets encompass specific ecosystem properties including condition (e.g., resilience), type (e.g., terrestrial), functions (e.g., carbon sequestration), and geographic dimensions (e.g., area). To achieve these targets, conservation actions must be informed by relevant spatial information (GBF 2022a, ECCC 2024). This information is needed to ensure decision-makers and practitioners can quantify, and make spatial predictions about, key ecosystem patterns , such as where different ecosystems occur and how their properties vary across space. Spatial models are essential for addressing these questions and ensuring targets are based on statistical evidence (Nicholson et al. 2019). Many spatial models will be built at the national or sub-national scale, as regional strategies are the main mechanism for implementing GBF resolutions (Perino et al. 2022, GBF 2022b). Yet spatial models established for predicting ecosystem patterns at regional extents are scarce (Geary et al. 2020, Naas et al. 2024), constraining plans to achieve GBF objectives.
Ecosystems are inherently complex (Riva et al. 2023). To help understand and reduce this complexity, ecologists formulate models to predict ecosystem patterns (Holling and Allen 2002). Spatial ecosystem patterns arise from differing combinations of biotic (e.g., taxa or traits) and abiotic (e.g., soil type) constituents , and fromaggregate properties (e.g., productivity, sequestered carbon) emerging from system development (Holling 1992, Artime and De Domenico 2022) (Figure 1). These patterns are shaped by the varied environmental circumstances within which ecosystems occur (Holling 1992). The rarity of spatial ecosystem models built for predicting these patterns (Geary et al. 2020) may stem from the very complexities that motivate modelling, as they present challenges for model design (Evans et al. 2013). More specific challenges include which ecosystem patterns andspatial scales should be selected for modelling, and why (Levin 1998, 2011, Gallagher et al. 2021). We present a general strategy for addressing these challenges with an alternative approach to ecosystem spatial pattern modelling. The patterns we model emerge from spatially structured biotic and abiotic ecosystem properties, predicted as a function of their shared relationships to environmental gradients. Our simultaneous modelling of both biotic and abiotic responses, including ecosystem structural, compositional, and functional variables, differs from the approaches taken in most spatial ecosystem models (see Geary et al. 2020).
Spatial ecosystem models are typically built to predict patterns of one or more ecosystem properties and those properties are often biotic (e.g., biomass, dispersal, decomposition) (Geary et al. 2020). In such models, abiotic variables are usually employed as predictors (van der Plas 2019, Hjort et al. 2022). For example, Huxley et al. (2023) show how topographic conditions shape linkages between biotic traits and primary productivity. Abiotic properties have less commonly been incorporated as response variables in spatial ecosystem models (Halvorsen et al. 2020). Here, abiotic properties are frequently limited to those chemical constituents (e.g., nutrients, carbon), with a direct link to biotic processes. Representative models (e.g., Lapierre et al. 2018, Soranno et al. 2019) predict how biotic (productivity) and abiotic (phosphorus, nitrogen) ecosystem properties vary with environmental drivers. Recent formulations (e.g., van der Plas et al. 2020, Gottschall et al. 2022) also predict biotic and abiotic properties, broadening their scope to include