The framework addresses a weakness in ecosystem service capacity matrices, which commonly assign one score to each combination of ecosystem type and service. Such matrices can incorporate expert judgment, statistical information and measurements, but often treat a land-cover class as homogeneous. Two forests assigned to the same category may differ in biomass, age structure, species composition or disturbance history, while grasslands at different stages of degradation may differ in their ability to support pollinators or limit erosion. The study defines ecosystem condition as the overall state of an ecosystem’s biophysical characteristics that underpin its capacity to generate services sustainably. Its central premise is that capacity scores should be modified when local condition differs from the typical state represented by the matrix.

To implement the adjustment, the authors separated ecosystem condition into structural and biological components. Structural condition captures physical and biophysical features such as vegetation cover, productivity, biomass and vertical organization. Biological condition encompasses biodiversity, species composition, functional diversity, conservation status and related features. The two axes can diverge: a plantation may have a closed canopy and substantial biomass but low biological condition, whereas a semi-natural grassland can support high floristic diversity with relatively little standing biomass. The framework therefore links each service to a functional group reflecting its combined dependence on structure and composition, rather than treating every service as controlled by a single condition component.

The authors created five functional groups. The first covers services primarily dependent on biomass and productivity, including carbon sequestration, erosion control and storm protection. The second includes services that depend on both ecosystem structure and specific composition, such as pollination, seed dispersal and wild plant resources. The third emphasizes biomass, plant cover and functional diversity, with timber production, flood regulation and agricultural production as examples. The fourth includes services strongly influenced by specific and functional composition, including recreation, aesthetic and habitat-related services. The fifth focuses on conservation status and biodiversity, including pest regulation, disease regulation and soil fertility maintenance. These groups are an internal methodological device, not a replacement for classifications of provisioning, regulating and cultural benefits.

Capacity scores came from a regional expert-elicited matrix for Hauts-de-France, which assigns ordinal values from 0 to 5 to combinations of ecosystem types and services. The medium-capacity matrix used the mean of the expert-score distribution, while the fifth and 95th percentiles were used to construct lower- and higher-capacity matrices. This is an important assumption rather than a direct observation that disagreement among experts represents ecological condition: the distribution of scores was treated as an envelope of plausible capacities for degraded and optimal states. The authors acknowledge that dedicated elicitation under explicitly low, medium and high condition scenarios would be methodologically preferable but substantially more resource-intensive.

Deterministic allocation tables translate condition into an adjusted score. Every 100-metre ecosystem unit is classified as low, medium or high in both structural and biological condition. For each functional group, the corresponding table identifies which capacity matrix to use, with the balance between structural and biological condition reflecting the service group’s ecological dependencies. Services such as carbon sequestration respond primarily to biomass and therefore place strong weight on structural condition. Pollination and wild plant resources require both suitable physical resources, such as flowering vegetation and nesting sites, and appropriate species assemblages. Pest regulation and other services in the fifth group depend more heavily on biological condition, although the allocation rules still consider both axes. The selected matrix supplies the relevant score, preserving the original 0–5 scale.

For the case study, the researchers built structural and biological condition layers at 100-metre resolution for 2020. The structural index combined Landsat tree-canopy cover, the 95th percentile of cloud-free annual Enhanced Vegetation Index observations and GEDI canopy-height data. The biological index combined the modelled Biodiversity Intactness Index, depth-weighted soil organic-carbon predictions from SoilGrids2.0 and the proportion of natural habitat within a 500-metre radius. The component variables were min–max normalized and combined with equal arithmetic weights because the study lacked empirical evidence for differential weighting, and the final indices were rescaled to a 0–1 range. The landscape was divided into aquatic, agricultural, forest and urban strata, with 20th- and 80th-percentile values defining low, medium and high condition within each ecosystem type. These pixel-level condition thresholds are distinct from the expert-score percentiles used to construct the low- and high-capacity matrices.

Threshold choices were evaluated using six reference datasets available across the study area: forest and soil carbon stocks for climate regulation, a habitat-suitability index for natural pest regulation, a model-based pollinator-suitability index for pollination, Strava running and cycling track density for nature-based recreation, a locally designed potential-production model for wild plant food resources, and interpolated standing wood volume for material and fibres. The researchers tested ten lower–upper percentile combinations and applied the adjustment procedure to every 100-metre pixel. For each service, Pearson correlations across pixels were calculated between the corresponding reference-layer values and the adjusted and unadjusted capacity estimates. The 20th–80th percentile scheme for both condition indices was selected for the principal maps, while the full comparison assessed how threshold choices affected performance.

Before adjustment, correlations ranged from r = 0.324 for pest regulation to r = 0.672 for climate regulation. With the selected ecosystem-specific thresholds, they rose to 0.454 and 0.708, respectively. The largest gain was for pest regulation, for which the correlation increased by 0.130, followed by wild plant resources, with a 0.103 increase. Climate regulation had the smallest improvement, at 0.036; the authors relate this to tree size and basal area already being partly represented by forest land-cover classification. Adjustment also changed the spatial pattern of estimates. Forest climate-regulation capacity ranged from 0.2 to 5.0 after adjustment, compared with 3.5 to 4.2 in the unadjusted matrix. Mean absolute differences were 0.73 for material and fibres, 0.81 for climate regulation and 0.58 for recreation, equivalent to 12–16% of the full 0–5 scale; the largest local difference for climate regulation reached 2.1, or 42% of the scale.

Ecosystem-specific rather than global thresholds prevented natural differences among ecosystem types from being mistaken for degradation. The 20th percentile of structural condition ranged from 0.16 in urban areas to 0.27 in forests, while the 80th percentile ranged from 0.43 to 0.90. A well-managed urban park could therefore receive a high structural-condition classification relative to other urban land even if its absolute canopy cover remained below that of a degraded forest. Adjustments were nevertheless service-specific. For material and fibres, most changes in forests were positive, indicating that the medium matrix underestimated capacity in above-average stands. Across the climate-regulation map, 45% of pixels decreased, 41% increased and 14% remained unchanged, with offsetting changes limiting the overall mean shift. Pairwise spatial concordance among the three mapped services was moderate, with Spearman correlations of 0.51–0.63, consistent with their differing dependence on structural and biological condition.

The study provides an operational method for connecting ecosystem-condition accounts with service-capacity matrices, but its evidence is geographically limited and the authors identify several constraints. The 20th–80th percentile thresholds were selected empirically in one landscape and require testing elsewhere. The deterministic tables do not fully represent interactions or nonlinear response curves between condition and capacity, and equal weighting combines variables operating over markedly different time scales, limiting sensitivity to short-term change. Other limitations include incomplete GEDI canopy-height coverage, the use of modelled rather than direct reference data for several services, and the restriction to biotic condition. The results support further validation against direct service measurements, empirical calibration of service-specific relationships, transferability tests in other regions, and incorporation of abiotic and spatially explicit landscape modifiers. Subject to those limitations, the case study shows that condition data can materially refine coarse ecosystem service maps without replacing them with a fully process-based model.

Subject of Research: Condition-adjusted estimation of ecosystem service capacity using spatial condition data and expert-scored matrices

Article Title: Condition Matters: Fine-Tuning Ecosystem Service Capacity with Ecosystem Condition

Article References: Roche, P., & Campagne, C. S. (2026). Condition Matters: Fine-Tuning Ecosystem Service Capacity with Ecosystem Condition. One Ecosystem, 11, Article e181068. https://doi.org/10.3897/oneeco.11.e181068

Image Credits: AI Generated

DOI: 10.3897/oneeco.11.e181068

Keywords: ecosystem services, ecosystem condition, capacity matrices, structural condition, biological condition, remote sensing, land cover, biodiversity, Scarpe-Escaut, ecosystem accounting, Condition, Matters