A quantitative approach to species occupancy across communities: the co-occurrence-occupancy curve

This paper introduces the Species Association Index (SAI), an occupancy-standardized metric derived from the co-occurrence-occupancy curve, to quantify and compare species' tendencies to associate with others across different communities while accounting for their overall frequency of occurrence.

The Gist

Imagine you are walking through a massive, bustling city. You see thousands of people going about their day. Some people are everywhere (like the barista at the corner coffee shop), while others are rare (like the person who only visits the library once a year).

Now, imagine you want to answer a simple question: Who hangs out with whom?

In the world of ecology, scientists have been trying to figure out which species of plants and animals live together in the same "neighborhoods" (habitats) and why. Usually, they try to guess this by looking at complex traits: "Do these two animals eat the same food?" or "Do they fight for space?"

But this new paper by Ontiveros and colleagues suggests a simpler, smarter way to look at the data. They introduce a new tool called the Co-occurrence-Occupancy Curve (or the "M-Curve") and a score called the Species Association Index (SAI).

Here is the breakdown in simple terms:

1. The Problem: The "Popular Kid" Bias

Imagine you are at a huge party.

• The Popular Kid: There is a guy named "Super-Common." He is at the party 99% of the time. Because he is there so often, he is likely to bump into everyone else, even if he doesn't actually like them.
• The Wallflower: There is a girl named "Rare." She only shows up 1% of the time. Even if she is a super-social person who loves everyone, she will rarely be seen with anyone else simply because she isn't there very often.

If you just count how many people they met, Super-Common looks like a social butterfly, and Rare looks like a loner. But that's misleading! You can't compare them fairly just by counting meetings. You need to know: Given that they are there, how likely are they to meet someone?

2. The Solution: The "M-Curve" (The Map of Expectations)

The authors created a graph (the M-Curve) that acts like a baseline expectation.

Think of it as a "Random Party Generator."

• If everyone at the party just wandered around randomly, ignoring each other and picking spots at random, what would the graph look like?
• The M-Curve shows exactly that. It draws a line showing: "If you are a species that shows up 50% of the time, you should, on average, meet X number of other species purely by chance."

This curve separates luck (being in the right place at the right time) from choice (actually wanting to be with someone).

3. The Score: The "Species Association Index" (SAI)

This is the paper's main invention. It's like a Z-score for social life.

Once you have the "Random Party" line (the M-Curve), you can look at any specific species and ask:

• Did they meet more people than the random line predicted?
  • Result: High SAI. They are "Social Butterflies." Maybe they are "ecosystem engineers" (like beavers building dams that create homes for others) or "epibionts" (like barnacles that love to hitch a ride on whales). They actively seek out company or create conditions where others can join them.
• Did they meet fewer people than the random line predicted?
  • Result: Low SAI. They are "Solitary Hermits." Maybe they are toxic, they create a hard shell that nothing else can live on, or they are so picky about their environment that no one else can survive there.

4. Real-World Examples from the Paper

The authors tested this on two very different "cities":

• The Mediterranean Rocky Shores (The Beach City):

  • They looked at plants and animals on rocks near the sea.
  • Finding: Most species fit the "Random Party" line perfectly. They just happened to be there.
  • The Outliers: Some species had very low scores. These were often species that created hard, crusty surfaces that were impossible for others to live on. They were the "loners" of the ocean. Others had high scores because they were mobile or opportunistic, hanging out with everyone.

• The Barro Colorado Island Rainforest (The Jungle City):

  • They looked at trees in a massive 50-hectare plot in Panama.
  • Finding: This forest is famous for being "neutral" (meaning trees don't really fight or help each other much; they just grow where they can). The data mostly fit the random line, which was expected.
  • The Twist: The few trees that did deviate from the line were linked to specific life strategies. Some trees were "growth-survival" trade-offs, meaning their specific way of growing determined who they lived next to.

Why Does This Matter?

Before this paper, if you saw two species living together, you might assume they were friends (or enemies) and start writing a complex story about their biology.

This paper says: "Wait a minute. Let's check the math first."

It gives ecologists a simple, transparent ruler to measure community life. It allows them to say, "Okay, this species is actually a loner, not because it's rare, but because it actively avoids others," or "This species is a social hub, bringing the whole community together."

In a nutshell:
The paper teaches us that frequency is not friendship. By using the M-Curve to account for how often a species shows up, and the SAI to measure how "social" it really is, we can finally see the true structure of nature's communities without getting fooled by the "popular kids" of the ecosystem.

Technical Summary

1. Problem Statement

Community ecology has traditionally relied on "top-down" approaches that require extensive data on species traits, interactions, and environmental filters to predict community structure. While powerful, these methods often obscure the causal mechanisms behind observed patterns. Conversely, "bottom-up" approaches using null models often struggle to distinguish between random stochasticity and genuine biological associations.

A specific, overlooked pattern exists: the relationship between a species' total occupancy (how frequently it appears across sites) and its co-occurrence strength (how many other species it shares sites with).

• The Bias: Common species naturally co-occur with more species simply because they are widespread, while rare species co-occur with fewer. This creates a confounding effect where raw co-occurrence counts are heavily biased by occupancy frequency.
• The Gap: There is a lack of a standardized, occupancy-independent metric to quantify a species' true propensity to associate with others, making it difficult to compare species with vastly different abundances or to detect deviations from neutral assembly processes.

2. Methodology

The authors propose a framework based on a bottom-up inference approach, utilizing a null model to establish a baseline expectation for species associations.

A. The Empirical Curve (M-curve)

• Definition: The authors define the M-curve (M for empirical), which plots the association strength ($A_i$) of a species against its global occupancy ($f_i$).
  • Occupancy ($f_i$): The proportion of sites where species $i$ is present ($N_i / N$).
  • Association Strength ($A_i$): The number of distinct species with which species $i$ co-occurs in at least one site within the metacommunity.
• Normalization: To allow cross-system comparison, the y-axis is normalized to relative association strength ($a_i = A_i / (S-1)$), where $S$ is the total number of species.

B. The Null Model

The authors derive the expected shape of the M-curve under a random null model with two core assumptions:

1. Species Independence: Species are placed across sites without influencing each other (no biotic interactions).
2. Site Equivalence: All sites are environmentally equivalent; species have no preference for specific sites.

Under these assumptions, the probability of two species co-occurring is derived using combinatorial probability (based on Veech, 2013). The expected number of co-occurrences for a focal species $i$ is calculated as the sum of probabilities of co-occurrence with all other $S-1$ species.

C. The Species Association Index (SAI)

To decouple association from occupancy, the authors introduce the Species Association Index (SAI), a z-score standardization:
$$s_i = \frac{A_i - \langle A_i \rangle}{\sqrt{\text{Var}(A_i)}}$$

• $A_i$: Observed co-occurrences.
• $\langle A_i \rangle$: Expected co-occurrences under the null model.
• $\text{Var}(A_i)$: Analytical variance of the co-occurrence distribution.
• Interpretation:
  • $s_i \approx 0$: The species follows the neutral expectation.
  • $s_i > 0$: The species co-occurs significantly more often than expected (positive association).
  • $s_i < 0$: The species co-occurs significantly less often than expected (negative association/segregation).

D. Theoretical Extensions

The authors extend the null model to account for:

1. Colonization-Extinction Dynamics: Modeling occupancy as a function of colonization ($c$) and extinction ($e$) rates.
2. General Occupancy Distributions: Using Beta distributions to model uneven species abundance (where most species are rare and few are common). They derive an analytical expression for the expected curve based on the occupancy distribution density function $F(p)$.

3. Key Contributions

1. Definition of the M-curve: Formalizing the relationship between occupancy and co-occurrence as a fundamental community-level pattern.
2. Development of the SAI: Creating a robust, occupancy-standardized metric that allows for the comparison of "sociability" across species with different frequencies.
3. Analytical Derivation: Providing closed-form mathematical solutions for the expected co-occurrence curve under random placement and specific dynamic models (colonization-extinction, Beta distributions), removing the need for computationally intensive randomization in many cases.
4. Benchmarking Neutrality: Establishing a transparent, testable baseline where neutrality is not an assumption but a null hypothesis against which real data can be compared.

4. Results

The framework was applied to two contrasting empirical systems:

A. Mediterranean Mediolittoral Habitats (MMH)

• Data: 143 rocky shore sites in Catalonia.
• Findings: Most species aligned with the null model. However, significant deviations were observed:
  • Low SAI: Species in simplified habitats (e.g., sandy shores) or those forming uncolonizable crusts showed lower co-occurrence than expected.
  • High SAI: Mobile species (gastropods), epibionts, and opportunists showed higher co-occurrence.
  • Boundary Detection: Species characteristic of different littoral zones (outside the mediolittoral fringe) showed the greatest deviations, suggesting the SAI can help define community boundaries and subunits.

B. Barro Colorado Island (BCI) Rainforest

• Data: 50-hectare forest plot with 1250 quadrats and 299 tree species.
• Findings: The system largely followed the neutral expectation, consistent with its history as a model for neutral theory.
• Drivers of Deviation: Species with high SAI variability were linked to specific demographic trade-offs:
  • Growth-Survival Trade-off.
  • Stature-Recruitment Trade-off: Specifically modulated by wood density ($p < 10^{-7}$).
  • This indicates that even in "neutral" systems, specific life-history strategies drive non-random co-occurrence patterns.

5. Significance and Implications

• Decoupling Frequency and Association: The SAI solves the "commonness bias," allowing ecologists to ask "who lives with whom" independent of "how common they are."
• Ecological Insight:
  • High SAI species tend to be structural/engineering species, epibionts, or opportunists in complex communities.
  • Low SAI species often generate simplified habitats or occupy environmentally impoverished patches.
• Methodological Rigor: The approach provides a statistically robust alternative to complex machine-learning models or trait-based approaches for initial community assembly screening. It highlights that departures from the M-curve are signatures of specific mechanisms (e.g., environmental filtering, biotic interactions, or historical contingency).
• Limitations: The method assumes a relatively homogeneous metacommunity. High environmental heterogeneity or mixing ecologically incomparable systems can undermine the "site equivalence" assumption, leading to false positives in deviation detection.

In conclusion, Ontiveros et al. provide a simple yet powerful quantitative tool to disentangle the effects of sampling scale and species frequency from true biological associations, offering a new lens for investigating the architecture of ecological communities.