Complementarity of ecosystem types drives landscape-wide productivity in North America
Data files
Feb 16, 2026 version files 16.62 MB
-
additive_partitioning_by_season.csv
8.28 MB
-
additive_partitioning.csv
2.49 MB
-
asynchrony.csv
92.57 KB
-
ecoregions.csv
416 B
-
README.md
8.71 KB
-
scaling.csv
298 B
-
spatio_temporal_partitioning.csv
21.28 KB
-
temporal_variation.csv
5.73 MB
Abstract
Landscape mosaics with a greater diversity of ecosystems tend to be more productive, mirroring the well-established relationship between species diversity and productivity observed in plot-scale biodiversity experiments. However, the mechanisms driving this effect at the landscape scale remain unclear. The data presented here was used to analyze a 15-year time series of satellite-derived primary productivity across over 50,000 landscape plots that vary in ecosystem type composition.
Our results demonstrate that more diverse landscapes are more productive and more predictable under environmental stress, especially drought. Using statistical partitioning, we show that these diversity effects are primarily driven by complementarity, with productivity gains that are broadly shared among ecosystem types rather than being dominated by a few. The specific ecosystem types that contributed most to landscape functioning varied regionally, but their role in driving mixture productivity remained unaffected by drought. These findings extend biodiversity theory to the landscape scale, emphasizing the critical role of higher-order diversity in shaping ecosystem function and informing landscape management.
Dataset DOI: 10.5061/dryad.r7sqv9srx
Description of the data and file structure
Landscape mosaics with a greater diversity of ecosystems tend to be more productive, mirroring the well-established relationship between species diversity and productivity observed in plot-scale biodiversity experiments. Here, we analyzed a 15-year time series of Landsat-satellite-derived primary productivity covering North America. Productivity was quantified as normalized difference vegetation index (NDVI), integrated over the growing season (NDVIgs, dimensionless). We established a network of over 50,000 plots ~25ha in size. These plots were chosen so that environmental conditions in these were decorrelated from land-cover (LC) type diversity, our variable of interest.
Mixed LC type plots were on average more productive than single LC type plots. We decomposed this effect of LC type diversity into complementarity and selection effects using the additive partitioning method by Loreau and Hector (2001) . We also partitioned these effects using an extension of the additive partitioning proposed by Isbell et al. (2018), who further considers spatial, temporal (inter-annual), and spatio-temporal components of selection effects. These spatio-temporal components are also known as 'insurance effects'. Temporal variation among years was considered by classifying years into dry and non-dry using a hydrological model; this classification predicted a large fraction of the observed inter-annual variance in productivity.
Files and variables
The diversity of a plot (ID) is quantified as land-cover (LC) type composition (column comp), indicating a combination of LC types, indicated by the letters A (agriculture), F (forest), G (grassland), S (shrubland), U (urban), and W (wetland). For example, "FG" indicates a mixture of the two LC types grassland and forest.
ecoregions.csv
| Column | Description |
|---|---|
| ecoregion | numeric ecoregion code |
| ecoregion_name | ecoregion name |
scaling.csv
| Column | Description |
|---|---|
| ecoregion | numeric ecoregion code |
| factor | factor |
additive_partitioning.csv
Additive partitioning components (complementarity effect, CE; selection effect, SE) calculated according to the partitioning scheme by Loreau & Hector (2001).
Note the the net diversity effect NE = CE + SE.
| Column | Description |
|---|---|
| ecoregion | numeric ecoregion code |
| block | block of study design |
| ID | study plot |
| comp | LC type composition |
| CE | complementarity effect |
| SE | selection effect |
additive_partitioning_by_season.csv
As additive_partitioning.csv, but calculated separately for the first quarter of the growing season ("start"), the last quarter of the growing season ("end"), and the main growing season in between ("center").
| Column | Description |
|---|---|
| ecoregion | numeric ecoregion code |
| block | block of study design |
| ID | study plot |
| comp | LC type composition |
| season | part of the growing season (start, center, end) |
| CE | complementarity effect |
| SE | selection effect |
spatio_temporal_partitioning.csv
Spatio-temporal partitioning according to Isbell et al. (2018).
Note that in the original paper, the spatio-temporal effects are
referred to as 'insurance effects'; here, we refer to them as
selection effect instead, since they are part of the total selection
effect.
dataset indicates the dataset the partitioning was applied to. See
manuscript for details.
- "overall": the variation across all plots is partitioned
- "within_ecoregions": the variation within within ecoregion is
partitioned - "across_ecoregions": using ecoregion averages, the variation between
ecoregions is partitioned
The following relationships apply (NE = net biodiversity effect):
- NE = TotalSE + TotalCE
- TotalSE = NonRandOY + AvgSE+TempSE + SpatSE + SpatTempSE
| Column | Description |
|---|---|
| dataset | dataset the partitioning was applied to |
| comp | LC type composition |
| TotalCE | total complementarity effect |
| TotalSE | total selection effect |
| TempSE | temporal selection effect |
| SpatSE | spatial selection effect |
| SpatTempSE | spatio-temporal selection effect |
| NonRandOY | non-random overyielding |
| AvgSE | average selection effect |
temporal_variation.csv
Interannual variation of productivity of each study plot, measured as
standard deviation of NDVI data.
| Column | Description |
|---|---|
| ecoregion | numeric ecoregion code |
| block | block of study design |
| ID | study plot |
| comp | LC type composition |
| NDVIgs.sd | inter-annual variation, quantified as standard deviation |
asynchrony.csv
Temporal variation and interannual asynchrony between different LC types.
Net diversity effects on temporal variation are calculated in two ways:
- delta.sd = sd.ab-0.5*(sd.a+sd.b) is the net effects not considering any
temporal asynchrony in productivity between component land-cover types in
the expected value expected under the null hypothesis. - delta.cov.sd = sd.ab-0.5sqrt(sd.a2 + sd.b2 + 2cov.mono) is the net effect
considering the ansynchrony between LC types, as observed in single LC
landscape plots in the same block (covariance stored in cov.mono).
| Column | Description |
|---|---|
| ecoregion | numeric ecoregion code |
| block | block of study design |
| comp | LC type composition |
| sd.a | inter-annual variation (as standard deviation) of first LC 'a' |
| sd.b | inter-annual variation (as standard deviation) of second LC 'b' |
| sd.ab | inter-annual variation (as standard deviation) of binary LC type composition 'ab' |
| cov.mono | inter-annual covariance of single LC plots with composition 'a' and 'b', in the same block |
| delta.sd | net diversity effect for composition 'ab', based on expectation that the productivity of the single LC type does not vary asynchronously |
| delta.cov.sd | net diversity effect, considering the asynchrony in productivity of the two LC types in the expected value |
References
- Loreau M, Hector A (2001) Partitioning selection and complementarity in biodiversity experiments. Nature 412, 72–76, https://doi.org/10.1038/35083573)
- Isbell F, Cowles J, Dee LE, Loreau M, Reich PB, Gonzalez A, Hector A, Schmid B (2018) Quantifying effects of biodiversity on ecosystem functioning across times and places^†^. Ecology Letters 21:763-778, https://doi.org/10.1111/ele.12928)
Code/software
All files are comma-separated UTF-8-encoded text files that can, for example, be read using the R software for statistical computing.
We used a network of over 50,000 study plots (area ~25 ha) spread across North America to relate landscape functioning to land-cover type diversity. For details of the study design, rationale, and method, we refer to https://dx.doi.org/10.1101/2025.11.17.688810.
In brief, to control for large-scale environmental variation, the plot network was blocked according to 16 ecoregions and 3° latitude ⨯ 6° longitude blocks. Land-cover type richness (LCR) was determined based on the Commission for Environmental Cooperation’s North American land monitoring system map for 2015 (30 m spatial resolution), with land-cover aggregated into the broad, ecologically distinct classes agriculture, forest, grassland, shrubland, wetland, and urban areas. Within each block, land-cover type richness gradients were created in such a way that these were not correlated with important topographic variables (e.g. slope and elevation) that likely also affected landscape productivity. Each plot's productivity was estimated as Normalized Difference Vegetation Index (NDVI), integrated over the growing seasons of the years 2008 to 2022.
Net diversity effects (NE) were calculated by subtracting single-land-cover plot values from the mixed-land-cover-plot data, as is common in biodiversity experiments. NE data was then statistically partitioned into complementarity and selection effects using the additive partitioning method by Loreau and Hector. Further, we applied the spatio-temporal partitioning method proposed by Isbell et al., using the classification of data into dry and non-dry years in lieu of calendar year, because the dry vs non-dry classification explained a large proportion of the observed temporal variation in NDVI.