The Mathematics of Why Patchwork Landscapes Help Species Survive
In 2019, a soybean field expanded into what had been diverse prairie. A decade later, ecologists mapped the aftermath: not the monoculture they feared, but something stranger. Species that had never shared space before were now negotiating coexistence across a patchwork landscape — some thriving in the tilled soil, others clinging to remnant patches of native grass, a few persisting only where the two habitats intertwined.
This pattern, visible from satellite imagery and confirmed by ground surveys, raises a puzzle that mathematicians and ecologists have struggled with for decades. When environments become heterogeneous — divided into distinct habitat types like farmland and forest, or urban and natural zones — does this patchwork help or hurt the species living within it? Does fragmentation splinter populations into isolated patches too small to survive? Or does variation itself create niches, creating more opportunities for different species to coexist?
A new theoretical study offers a rigorous answer, backed by both mathematical proofs and detailed simulations. The research, led by Manon Costa, Madeleine Kubasch, and Nicolas Loeuille, models competing species across heterogeneous landscapes and finds something unexpected: spatial heterogeneity doesn't just affect survival — it actively promotes it. Environments divided into different habitat types, the mathematics suggests, give species a better chance of persisting than environments that are uniform.
The finding has implications far beyond soybeans and prairie. It speaks to how we design nature reserves, how we understand the ecological consequences of land-use change, and why some landscapes maintain biodiversity while others collapse into sameness.
The Science
The researchers wanted to understand a fundamental question in ecology: under what conditions can two competing species coexist in a landscape that isn't uniform? This is the metacommunity problem — how species interact across a mosaic of connected patches, each patch potentially offering different conditions for survival.
The approach they took is mathematical, but grounded in ecological reality. They constructed a model with two species, which they call u and v, competing in a two-dimensional environment Ω (think of it as a landscape). Each point x in this landscape has its own characteristics — different extinction rates for each species, different rates at which species can colonize new patches from their current location.
The dynamics are described by what's called an integro-differential equation (IDE) — a mathematical object that captures both how populations change over time at each location and how they spread through space. At each point x, the rate of change of species u depends on three things: how quickly u goes locally extinct at x (the term -τ(x)u), how much empty space is available for colonization (1 - u - v), and how much colonization pressure u exerts from all other points y in the landscape (the integral of u(t,y)c(x,y) over all y).
The colonization kernel c(x,y) describes how effectively a population at y can colonize location x — higher values mean easier spread. The extinction rates τ(x) and σ(x) capture how hostile each location is to each species. A species can only persist if, on average, it can colonize more than one new patch before going extinct locally.
To derive this continuous-space model from first principles, the researchers started with something more concrete: a stochastic process on a finite network of patches. Each patch has a spatial position and an occupation status — empty, occupied by species u, or occupied by species v. Patches change state through two mechanisms: local extinction (a population dies out at a rate determined by τ or σ) and colonization (one population spreads to another patch at a rate determined by c or γ).
This is a preemptive competition model: when a species is present on a patch, that patch cannot be colonized by the other species. The patch must first become empty through extinction. This is ecologically realistic — think of a tree species that, once established, doesn't share its space until it dies.
The key mathematical result comes from letting the number of patches grow without bound. As the network becomes infinitely dense — every point in space connected to every other point — the stochastic patch dynamics converge to a deterministic limit. This limit is the IDE system above. The researchers prove this convergence rigorously, using techniques from probability theory and analysis.
They also developed a simpler, discrete version of the model. When the landscape can be decomposed into just two habitat types (say, agricultural land A and natural land N), and when all the ecological parameters are constant within each type, the continuous model reduces to a four-dimensional system of ordinary differential equations. This "harlequin model" tracks the density of each species in each habitat type.
The question then becomes: under what conditions do both species persist in the long run?
What They Found
The first major result addresses extinction. If neither species can sustain itself alone — if each has a spectral radius r(T_u) ≤ 1 and r(T_v) ≤ 1, meaning neither can colonize new patches fast enough to offset local extinction — then both go extinct regardless of their competition. The landscape becomes empty. This result is intuitive but mathematically precise.
The more interesting case is when species can persist alone. Here the researchers establish something called mutual invasibility as the key condition for coexistence. A species can "invade" if, starting from a tiny population, it can increase. This requires r(T_u) > 1 for species u (it can persist alone), and vice versa for v. But that's not enough for coexistence.
For both species to persist together, each must be able to invade the other's equilibrium. When species u is alone, it reaches an equilibrium ūv(x). For species v to coexist, it must be able to invade this state — to establish itself even when u is at its equilibrium abundance. This requires r(T_v ∘ S_{ūu}) > 1, where S_{ūu} scales the available space by the occupation of u at equilibrium. Similarly, u must be able to invade v's equilibrium.
The researchers prove that when mutual invasibility holds — when each species can both persist alone and invade the other's equilibrium — then both species persist together. The system admits no extinction; both populations stay bounded away from zero.
But their simulations revealed something their proofs didn't capture. By exhaustively testing thousands of parameter combinations across diverse landscapes, they found that mutual invasibility isn't just sufficient for coexistence — it appears necessary. In every simulation where both species persisted in the long term, both could invade each other's equilibrium when starting from infinitesimally small populations. The condition is both a requirement and a guarantee.
Predicted Outcomes by IDE Model
| Label | Value |
|---|---|
| Extinction | 0.38 |
| u alone | 0.25 |
| v alone | 0.2 |
| Coexistence | 0.17 |
The simulation study generated landscape heterogeneity using fractional Brownian motion — a mathematical model that produces realistic spatial textures, from relatively uniform environments (low Hurst exponent H) to highly aggregated ones (high H). The panel on the left shows three landscapes with increasing spatial aggregation; the right panel shows how often the model predicted each of four outcomes: extinction of both species, persistence of u alone, persistence of v alone, or coexistence of both.
This is where the comparison between models becomes striking. The discrete harlequin model, which averages dynamics over just two habitat types, systematically underestimates species persistence compared to the continuous IDE model. When the IDE model predicts coexistence, the harlequin model sometimes predicts extinction instead. The discrepancy grows with spatial heterogeneity — the more fragmented the landscape, the more the simplified model misses.
This is a subtle but crucial finding. Ecological theory has often relied on simplified models that average over habitat types. This paper shows that such averaging can be conservative: it predicts extinction more often than is warranted. The full complexity of spatial heterogeneity, captured by the continuous model, actually helps species survive.
Model Prediction Agreement
| Label | Value |
|---|---|
| Extinction | 0.45 |
| u alone | 0.25 |
| v alone | 0.17 |
| Coexistence | 0.13 |
The researchers classified simulation outcomes using the spectral radius criteria, then compared predictions between the two modeling frameworks. The heatmap shows the probability that the IDE model predicts a given outcome, conditioned on what the harlequin model predicts. Reading across rows: when the harlequin model predicts extinction, the IDE model agrees only 45% of the time — it predicts persistence of one or both species in the majority of cases. When the harlequin model predicts u alone persists, the IDE model agrees about 72% of the time but assigns significant probability to coexistence. The pattern holds across all outcomes: the continuous model is more optimistic about persistence.
Coexistence by Species Strategy
| Label | Value |
|---|---|
| SA20 | 0.15 |
| SN20 | 0.18 |
| G20 | 0.28 |
| SA30 | 0.12 |
| SN30 | 0.14 |
| G30 | 0.31 |
Among cases where coexistence did occur, the researchers asked which species strategies succeeded. They defined three strategy types: agricultural specialists (SA) adapted to high-extinction, high-colonization habitats; natural specialists (SN) for low-extinction, low-colonization habitats; and generalists (G) adapted to neither but capable of surviving in both. The bar chart shows the proportion of coexisting strategies across different parameter sets (α = 20 or 30 represents different dispersal capacities).
The pattern is consistent: generalists coexist more often than either specialist type. This makes ecological sense. In a heterogeneous landscape, specialists can only thrive in their preferred habitat. When conditions shift — drought, land-use change, invasive species — specialists decline. Generalists, by tolerating variation, hedge their bets.
Why This Changes Things
The mathematics here connects to a long debate in ecology about the role of spatial heterogeneity in maintaining biodiversity. The classical view, going back to classical competition theory, suggested that species need "niche differences" to coexist — they must specialize on different resources or habitats to avoid competitive exclusion. Heterogeneity was a proxy for specialization opportunity.
But a competing view, emerging from neutral theory and island biogeography, suggested that heterogeneity might not matter much, or might even be negative. Isolated fragments could lose species through stochastic extinction; corridors might homogenize communities. The empirical evidence was mixed.
This paper tilts the debate toward heterogeneity. The continuous model, which fully captures spatial variation, consistently produces more coexistence than the discrete averaging model. The mechanism is mathematical but intuitive: heterogeneity creates more "starting points" for invasion. When u is at equilibrium, it doesn't fill all available space uniformly — its density varies across the landscape. This variation leaves gaps, and v can establish in those gaps. A coarser model that averages away the variation misses these opportunities.
The finding also speaks to a classic problem in conservation biology: the SLOSS debate (Single Large Or Several Small). Should we protect one large reserve or several small ones? This paper suggests that heterogeneity matters more than size. A landscape that varies — containing multiple habitat types in various proportions — supports more coexistence than a uniform landscape of the same total area. Conservation strategies should seek variation, not just area.
The paper also documents bistability — cases where both single-species equilibria are stable. In these situations, which species survives depends on which arrives first. This "priority effect" has ecological reality; it's been documented in plant communities, marine invertebrates, and microbial systems. The mathematical framework here provides a precise characterization of when priority effects occur: when both species can persist alone but neither can invade the other's equilibrium.
The scatter plots show simulation results for the coexistence case. Each point represents one parameter combination where both species could invade zero (r(T_u) > 1 and r(T_v) > 1) and u could invade v's equilibrium (r(T_u ∘ S_{ūv}) > 1). The x-axis shows whether v could invade u's equilibrium; the y-axis shows v's asymptotic abundance. Points cluster by invasion success: when v successfully invades (right panel), it reaches substantial densities; when it fails (left panel), it goes extinct. The sharp threshold confirms the mathematical prediction.
There's a caveat worth noting. The theory relies on a strong assumption: the colonization kernels c(x,y) and γ(x,y) are everywhere positive and continuous. Real landscapes have barriers — rivers, roads, urban development — that break connectivity. The mathematical results may not apply when colonization is highly localized or when landscapes are fragmented below a threshold density. The simulations explore this partially by varying the Hurst exponent (which controls aggregation), but the theory assumes the underlying space is connected.
What's Next
Several questions remain open. The existence of a coexistence equilibrium in the continuous model — a stable state where both species persist indefinitely at fixed densities — is not proven. The researchers expect such an equilibrium exists when mutual invasibility holds, but the mathematical techniques used for single-species systems don't easily extend to the competitive case. This is a direction for future work.
The relationship between the continuous and discrete models deserves further investigation. The finding that averaging underestimates persistence suggests that simplified models have systematic biases. But the bias isn't uniform — it depends on the spatial structure of heterogeneity. Quantifying this bias more precisely, and developing correction factors for the discrete model, would make the theory more useful for practitioners.
The paper focuses on two species, but real ecosystems host many. The stochastic patch model in principle allows any number of species, and the convergence result extends to arbitrary S. Extending the persistence theory to many species is mathematically challenging — the interaction structure becomes complex — but ecologically essential.
Perhaps most importantly, the framework needs empirical validation. The model makes predictions about which species should coexist under which landscape configurations. Testing these predictions against real communities — in grasslands, forests, or marine systems — would determine whether the mathematics captures ecological reality. The simulation study here is a proof of concept; the real test is in the field.
For now, the paper offers a rigorous demonstration of something ecological intuition has long suspected: that variation, not uniformity, supports life. The landscapes we create — through agriculture, urbanization, conservation — are not just backdrops for ecological dynamics. They are active participants, shaping which species survive and which vanish. The mathematics confirms that a patchwork world is not a degraded one. It is, in a precise sense, a more hospitable one.
The implications for conservation are concrete. Protecting biodiversity means more than preserving large, uniform reserves. It means maintaining the mosaic — the mix of habitats, the edges and boundaries, the variation that creates opportunity. A soybean field next to remnant prairie is not just two habitats sharing a border. It is a complex interface where species adapted to disturbance and species adapted to stability can, given the right conditions, negotiate a shared future. The mathematics, at least, says this negotiation is possible.