Biodiversity and Ecosystem Function

In a decade of yearly harvests at Cedar Creek, Minnesota, grassland plots sown with more plant species varied less in total production, while the individual species inside them varied more. That result raises the question this lesson answers: when species are lost, does what an ecosystem does change, and if so, why?

The question: does losing species change what an ecosystem does?

An ecosystem function is a rate or stock of a process: how much plant biomass a field produces in a year (its primary productivity, the subject of energy flow), how much nitrogen it holds, how fast litter decays. Species richness is the number of species present, the simplest diversity measure from diversity and succession.

The obvious way to ask whether richness matters is to survey real grasslands and plot biomass against species number. Surveys often find a positive relationship. They cannot tell you why. A wetter or more fertile patch could support both more species and more biomass, so richness and biomass would rise together without one causing the other. That third variable is a confounder.

So ecologists built communities instead of finding them.

Building communities: Cedar Creek and BIODEPTH

At Cedar Creek, Tilman and colleagues seeded 168 plots, each 9 by 9 metres, in spring 1994. Each plot was sown with 1, 2, 4, 8 or 16 perennial grassland and savanna species, chosen at random from a pool of 18. Every plot received the same seed mass, 10 g per square metre in 1994 and 5 g in 1995, so rich plots did not simply get more seed. Aboveground biomass was measured each year by clipping vegetation in sampling strips, drying it and weighing it.

Random composition is the key move. If you hand-picked the species for each richness level, any richness effect might be the effect of the species you picked. Drawing many random compositions at each level means a consistent difference between levels belongs to richness itself.

BIODEPTH repeated the logic at eight European grassland sites, each using local species. Hector and colleagues reported in 1999 that, across sites, losing plant species gave an overall log-linear decline in average aboveground biomass. Composition still mattered: at a fixed species number, communities with fewer functional groups (such as grasses, legumes and other herbs) were less productive.

The approach has since been taken to trees. In the BEF-China experiment, a planted subtropical forest, Huang and colleagues reported in 2018 that 16-species mixtures had accumulated more than twice the carbon of the average monoculture after eight years, and an amount similar to two commercial monocultures. The benchmark matters, as the next section shows.

Complementarity versus selection: splitting the extra yield

A mixture can outyield its monocultures for two quite different reasons. Think of a relay team. It can win because each runner covers the leg they are good at (complementarity), or because a random team of four is more likely to include one superb runner who carries the race (selection; a positive selection effect is what the sampling-effect hypothesis predicts). In plants, complementarity covers resource partitioning, the niche differences you met in competition, and positive interactions such as legumes adding nitrogen. Selection covers dominance by species with particular traits.

Loreau and Hector’s 2001 partition measures both. For each species, the relative yield RY is its yield in mixture, Y, divided by its monoculture yield, M. The expected relative yield is usually its sown proportion. The change, ΔRY, is observed minus expected. With N species:

net effect = complementarity + selection complementarity = N x mean(ΔRY) x mean(M) selection = N x cov(M, ΔRY), using the population covariance (divide by N)

Worked problem. In this constructed data set, two species are sown in equal proportions. Monoculture yields are M_A = 400 and M_B = 100 g of dry biomass per square metre. In the mixture, Y_A = 300 and Y_B = 100 g per square metre. Assume the yields are measured on plots of the same area in the same year.

1. Net effect. Expected mixture yield is the mean monoculture yield, (400 + 100)/2 = 250 g per square metre. Observed mixture yield is 300 + 100 = 400. Net effect = 400 - 250 = +150 g per square metre.
2. Relative yields. RY_A = 300/400 = 0.75; RY_B = 100/100 = 1.00. Each was sown at half, so expected RY is 0.5. ΔRY_A = 0.25 and ΔRY_B = 0.50, mean 0.375. Mean M = 250.
3. Complementarity. 2 x 0.375 x 250 = +187.5 g per square metre.
4. Selection. Deviations from the means: species A has M - 250 = +150 and ΔRY - 0.375 = -0.125; species B has -150 and +0.125. The products are -18.75 and -18.75, and their mean (divide by 2) is -18.75. Selection = 2 x -18.75 = -37.5 g per square metre.
5. Check. 187.5 - 37.5 = 150, the net effect.
6. Against the best monoculture. The mixture yields 400, exactly what species A yields alone. There is no transgressive overyielding: the mixture beats the average monoculture but not the best.

Read the signs. Both species did better than expected, so complementarity is positive. But the gain was larger for the low-yielding species B, so selection is negative.

Monoculture yield M_A 400 g m⁻² Monoculture yield M_B 100 g m⁻² Mixture yield Y_A 300 g m⁻² Mixture yield Y_B 100 g m⁻²

Expected yield (mean M)

250 g m⁻²

Observed yield (Y_A + Y_B)

400 g m⁻²

Relative yields RY_A, RY_B

0.75, 1.00

Net effect

+150.0 g m⁻²

Complementarity = 2 × mean(ΔRY) × mean(M)

+187.5 g m⁻²

Selection = 2 × population covariance

−37.5 g m⁻²

Against the best monoculture

400 vs 400 g m⁻²

Loading.

In this sketch: two species sown in equal proportions (expected RY = 0.5 each), plots of the same area in the same year, constructed yields.

Net effect = complementarity + selection, and the split is algebra on yields: it does not say which biological process produced them. Beating the average monoculture is overyielding; beating the best one is transgressive overyielding.

In BIODEPTH, the complementarity effect was positive overall, while the selection effect averaged zero and ranged from negative to positive among sites. At Cedar Creek, Fargione and colleagues found that over 1996 to 2005 the balance shifted from sampling toward complementarity, with selection strongly negative by the end.

A positive complementarity term does not prove resource partitioning. The partition is algebra on yields; it does not identify which biological process produced them, and it needs species-level yields in both mixtures and monocultures.

Stability: resistance, resilience and averaging

“Stable” has several meanings, and they need not agree.

• Temporal stability is the mean of total community biomass across years divided by its standard deviation (μ/σ).
• Resistance is how little a function changes during a disturbance such as a drought.
• Resilience is how fully or quickly it returns afterwards.

Why might richness raise temporal stability? Think of an investment portfolio. If species fluctuate independently, their ups and downs partly cancel in the total; this is statistical averaging. If species fluctuate in opposite directions, one rising as another falls (negative covariances), the cancelling is stronger; that is compensatory dynamics.

To see averaging in made-up numbers, suppose each species has a mean of 50 g per square metre and a standard deviation of 25, and fluctuations are independent. One species gives μ/σ = 50/25 = 2. Four species give a mean of 200 and a standard deviation of the square root of 4 x 25^2, which is 50, so μ/σ = 4. Under these assumptions, stability rises with the square root of species number, with no species becoming any steadier.

Tilman, Reich and Knops reported that at Cedar Creek over a decade, greater richness increased the temporal stability of annual aboveground production even though individual species were less stable, and they found no evidence of a covariance effect. Zhao and colleagues (2022) found statistical averaging was the principal mediator in their North American grassland analysis. Luo and colleagues (2025) showed that synchrony patterns can reverse with the length of the time series, so the inferred mechanism depends on how long you watch.

Resistance and resilience separate. Across 46 grassland diversity experiments, Isbell and colleagues (2015) found that productivity in communities of one or two species changed by about 50 percent during wet or dry climate events, against about 25 percent at 16 to 32 species. By one year after each event, they detected no dependence of recovery on diversity. And in 690 aquatic ciliate microcosms, Pennekamp and colleagues (2018) found that richness raised temporal stability but lowered resistance to warming. Diversity does not raise every kind of stability.

Field evidence and confounding

Experiments assign richness; real landscapes do not. Duffy and colleagues (2017) synthesized 67 field studies of wild systems, 133 estimates in all, and found positive biodiversity-biomass associations after statistically adjusting for environmental covariates. That brings the question out of small plots, but adjustment is not random assignment: an unmeasured variable could still drive both. These results carry forward into global change and conservation.

Design problem

An observational survey finds that grassland plots with more plant species have more aboveground biomass. Design a study to decide whether richness causes the difference.

One way to do it

Start with the confounder: soil fertility or moisture could raise both richness and biomass. Break the link by assigning richness yourself.

1. Divide a uniform field into blocks by soil and slope. Within each block, randomly assign plots to 1, 2, 4 or 8 species.
2. At each richness level, draw several independent random compositions from a fixed local pool, and include a monoculture of every species so the partition can be computed.
3. Keep plot area and total seed mass constant across treatments, and weed every plot the same way to hold its assigned composition.
4. Measure soil nitrogen, moisture and light in every plot before sowing, and check they do not differ by treatment.
5. Cross richness with a randomly assigned nitrogen or drought treatment, so you can see whether the richness effect holds within each environment.
6. Treat the plot as the unit, harvest species separately each year for several years, and analyse assigned richness, with block as a factor.
7. Report net, complementarity and selection effects alongside raw yields, compare mixtures with the best monoculture as well as the average, and predefine the stability measures (μ/σ over a fixed window, resistance and recovery for any event).