Understanding Disease Spread Through Space:
The SIR Model with Nonlocal Mobility

A Comprehensive Introduction to Spatial Epidemic Modeling with Kernel-Based Movement

Epidemiological PDE Model Toolbox Series
October 2025

Abstract: The classical SIR (Susceptible-Infected-Recovered) model has been fundamental to understanding epidemic dynamics, but its standard formulation assumes well-mixed populations. Real-world disease transmission, however, occurs in spatially structured populations where individuals move across landscapes. The SIR model with nonlocal mobility introduces spatial kernels to capture realistic movement patterns, ranging from local diffusion to long-distance jumps. This article provides a comprehensive introduction to this spatially explicit framework, explaining its mathematical formulation, biological interpretation, parameter estimation, and modern extensions that have proven invaluable for modeling diseases from influenza to COVID-19.

🦠 1. Introduction: Why Space Matters in Epidemics

When the COVID-19 pandemic swept across the globe in early 2020, epidemiologists quickly recognized that understanding where people moved was just as important as understanding how the disease transmitted. Traditional epidemic models often treat populations as perfectly mixed—imagine everyone in a city randomly bumping into everyone else with equal probability. While this assumption simplifies mathematics, it fails to capture crucial spatial patterns that determine real outbreak dynamics.

Consider how diseases actually spread: a person infected in New York might fly to Los Angeles, seeding a new outbreak cluster thousands of kilometers away. Meanwhile, their neighbors back home face higher infection risks due to proximity. This combination of local transmission and long-distance mobility creates complex spatial patterns that classical models cannot capture .

The SIR model with nonlocal mobility elegantly addresses this limitation by incorporating movement kernels—mathematical functions describing how likely individuals are to move different distances. This framework has revolutionized spatial epidemiology, enabling researchers to predict disease spread across cities, countries, and continents with unprecedented accuracy.

📊 2. Model Description: The Mathematics of Spatial Disease Spread

2.1 The Classical Foundation

Before introducing spatial complexity, let’s revisit the classical SIR framework. At any time t, a population is divided into three compartments:

S(t) = Susceptible individuals (can contract the disease)

I(t) = Infected individuals (currently sick and infectious)

R(t) = Recovered individuals (immune or deceased)

The classical model assumes disease transmission follows the “law of mass action,” where infection rate depends on contact frequency between susceptible and infected individuals .

2.2 Adding Space: The Nonlocal Mobility Framework

Now we extend this to space. Let x represent spatial location (say, coordinates on a map), and t represent time. Our model describes how disease spreads across a landscape using partial differential equations (PDEs) with integral terms capturing movement.

∂S(x,t)/∂t = ∫ K(x−y)[S(y,t) − S(x,t)]dy − βS(x,t)I(x,t)

∂I(x,t)/∂t = ∫ K(x−y)[I(y,t) − I(x,t)]dy + βS(x,t)I(x,t) − γI(x,t)

∂R(x,t)/∂t = ∫ K(x−y)[R(y,t) − R(x,t)]dy + γI(x,t)

🔍 Understanding the Equations: Each equation has two key components. The integral term (∫…) represents movement through space, while the remaining terms capture disease dynamics identical to the classical model.

2.3 The Mobility Kernel: K(x−y)

The heart of this model is the mobility kernel K(x−y), which describes movement probability between locations. Think of it as a “movement landscape”—higher values indicate more frequent travel between those locations.

💡 Key Insight: The integral ∫ K(x−y)[S(y,t) − S(x,t)]dy means: “susceptible individuals arrive at location x from all other locations y, weighted by the kernel K, while simultaneously departing from x to other locations.”

Common Kernel Forms:

Gaussian Kernel (most popular): K(z) = (1/(4πD))exp(−|z|²/(4D))

Models normal diffusion with variance parameter D. Most movement is local, with rare long-distance events.

Exponential Kernel: K(z) = (1/(2πα²))exp(−|z|/α)

Produces heavier tails—more long-distance movement than Gaussian, useful for air travel patterns.

Power-law Kernel: K(z) = C/(|z|^μ + 1)

Captures scale-free movement observed in human mobility data (think cell phone tracking studies) .

⚙️ 3. Parameter Definitions and Biological Interpretation

Parameter	Description	Typical Values	Units
β	Transmission rate (infection probability per contact)	0.2–2.0 day−1	1/(person·day)
γ	Recovery rate (inverse of infectious period)	0.1–0.5 day−1	1/day
D	Diffusion coefficient (mobility intensity)	0.01–10 km²/day	length²/time
α	Kernel characteristic length scale	1–100 km	length
R0	Basic reproduction number = β/γ	1.5–6.0	dimensionless

🎯 Parameter Estimation in Practice

Transmission rate (β): Estimated from early epidemic growth rates or contact tracing data. For COVID-19, typical values ranged from 0.3–0.7 day⁻¹ depending on interventions.

Recovery rate (γ): Inverse of the mean infectious period. For influenza: γ ≈ 0.33 day⁻¹ (3-day infectious period). For measles: γ ≈ 0.1 day⁻¹ (10-day period).

Mobility parameters: Estimated from GPS data, mobile phone records, or transportation statistics. Studies show human movement often follows power-laws with exponents around μ = 2–3.

🌍 4. Assumptions and Applicability

4.1 Core Assumptions

1. Homogeneous mixing at each location: The model assumes within any small spatial region, individuals mix randomly. Valid for cities but may break down in highly segregated populations.

2. Movement independent of disease status: Individuals move according to kernel K regardless of whether they’re susceptible, infected, or recovered. Violated if sick individuals self-isolate or avoid travel.

3. Constant population: No births, deaths (except disease-induced), or immigration. Reasonable for fast epidemics (weeks to months) but invalid for endemic diseases over years.

4. Spatial homogeneity in parameters: β and γ are typically assumed constant across space. Can be relaxed by making β(x) and γ(x) spatially varying.

4.2 When to Use This Model

✅ Ideal Scenarios:

• Acute outbreaks where spatial spread is crucial (COVID-19, Zika, Ebola)
• Regional to continental scales (10–10,000 km)
• Populations with measurable mobility patterns
• Questions about spatial control strategies (where to vaccinate, close borders, etc.)

❌ Less Suitable:

• Very local scales (single buildings) where stochastic effects dominate
• Diseases with complex within-host dynamics requiring structured models
• Scenarios with strong behavioral responses to infection (panic, evacuation)

🔬 5. Model Extensions and Variants

5.1 SEIR with Nonlocal Mobility

Many diseases have a latent period where individuals are infected but not yet infectious. The SEIR variant adds an “Exposed” compartment:

∂E(x,t)/∂t = ∫ K(x−y)[E(y,t) − E(x,t)]dy + βS(x,t)I(x,t) − σE(x,t)

where σ is the rate of progression from exposed to infectious (1/σ = incubation period). Application: Essential for COVID-19 modeling, where the ~5-day incubation period strongly affects control timing.

5.2 Multi-Patch Models with Network Kernels

Instead of continuous space, divide the region into discrete patches (cities, counties) connected by a network. The kernel becomes a matrix K_ij representing travel from patch i to j :

dI_i/dt = Σ_j K_ij(I_j − I_i) + βS_iI_i − γI_i

Purpose: Computationally efficient for large regions. Application: Global airline network models for pandemic influenza, using flight passenger data to define K_ij.

5.3 Asymmetric and Anisotropic Kernels

Real movement isn’t always symmetric. Mountains, rivers, or transportation infrastructure create directional biases:

K(x, y) = K₀exp(−[(x − y)^TΣ⁻¹(x − y)]^1/2)

where Σ is a covariance matrix encoding preferred movement directions. Application: Modeling disease spread along river valleys or transportation corridors.

5.4 State-Dependent Mobility

Relax the assumption that movement is disease-independent. Infected individuals might reduce mobility :

∂I(x,t)/∂t = ∫ θK(x−y)[I(y,t) − I(x,t)]dy + βS(x,t)I(x,t) − γI(x,t)

where 0 ≤ θ ≤ 1 represents reduced mobility of infected individuals. Application: Diseases with severe symptoms (Ebola) where sick people naturally travel less.

5.5 Age-Structured Spatial Models

Combine spatial kernels with age stratification to capture age-dependent contact patterns and mobility:

∂I_a(x,t)/∂t = ∫ K_a(x−y)[I_a(y,t) − I_a(x,t)]dy + S_a(x,t)Σ_bβ_abI_b(x,t) − γI_a(x,t)

where subscript a denotes age group, and β_ab is transmission rate from age b to age a. Children move less than adults (K_child more peaked than K_adult). Application: Optimizing age-targeted vaccination in spatially structured populations.

5.6 Time-Varying Kernels for Intervention Effects

Lockdowns and travel restrictions change mobility patterns. Model this with time-dependent kernels:

K(z, t) = K₀(z)·m(t)

where m(t) is a time-varying mobility reduction factor (e.g., m(t) = 0.2 during strict lockdown means 80% reduction). Application: Evaluating COVID-19 stay-at-home orders using cell phone mobility data.

💻 6. Numerical Implementation and Computational Considerations

Solving these nonlocal PDE models requires specialized numerical methods:

Spatial Discretization: Divide space into grid cells or use finite element meshes. The integral kernel becomes a convolution that can be computed efficiently using Fast Fourier Transforms (FFT).

Time Stepping: Explicit schemes (Euler, Runge-Kutta) are simple but require small time steps. Implicit-Explicit (IMEX) schemes handle the stiff reaction terms more efficiently.

Computational Cost: For N spatial points, naive kernel evaluation is O(N²) per time step. FFT-based convolutions reduce this to O(N log N), enabling continental-scale simulations.

🎯 7. Conclusion: The Future of Spatial Epidemic Modeling

The SIR model with nonlocal mobility represents a powerful synthesis of classical epidemiology and modern spatial analysis. By explicitly incorporating realistic movement patterns through flexible kernel functions, this framework has become indispensable for understanding and predicting disease spread in our interconnected world.

Recent advances continue to refine these models: machine learning techniques now extract mobility kernels directly from big data sources, multi-scale approaches link individual movement to population-level patterns , and agent-based implementations capture stochasticity while maintaining spatial realism .

As we face emerging infectious diseases and potential future pandemics, spatial epidemic models will play an increasingly critical role in public health planning. Understanding how diseases move through space—captured mathematically through elegant kernel formulations—remains one of our most powerful tools for protecting human health.

🌟 Key Takeaway: The beauty of the nonlocal mobility framework lies in its flexibility. By choosing appropriate kernels, the same mathematical structure can describe local diffusion, long-distance airline transmission, or complex network patterns—making it a truly universal language for spatial epidemiology.

📚 References

[1] Brockmann, D., & Helbing, D. (2013). The hidden geometry of complex, network-driven contagion phenomena. Science, 342(6164), 1337-1342. https://www.science.org/doi/10.1126/science.1245200

[2] Balcan, D., et al. (2009). Multiscale mobility networks and the spatial spreading of infectious diseases. Proceedings of the National Academy of Sciences, 106(51), 21484-21489. https://www.pnas.org/doi/10.1073/pnas.0906910106

[3] Berestycki, H., Roquejoffre, J. M., & Rossi, L. (2016). The influence of a line with fast diffusion on Fisher-KPP propagation. Journal de Mathématiques Pures et Appliquées, 106(5), 743-785.

[4] Kraemer, M. U., et al. (2020). The effect of human mobility and control measures on the COVID-19 epidemic in China. Science, 368(6490), 493-497. https://www.science.org/doi/10.1126/science.abb4218

[5] Kermack, W. O., & McKendrick, A. G. (1927). A contribution to the mathematical theory of epidemics. Proceedings of the Royal Society A, 115(772), 700-721. https://royalsocietypublishing.org/doi/10.1098/rspa.1927.0118

[6] Ruan, S. (2007). Spatial-temporal dynamics in nonlocal epidemiological models. In Mathematics for Life Science and Medicine (pp. 97-122). Springer. https://link.springer.com/chapter/10.1007/978-3-540-34426-1_5