Imagine an urban atmosphere filled with two distinct types of airborne particles: pure black carbon (soot) emitted from a diesel truck, and pure ammonium sulfate emitted from a coal-fired power plant. If these particles remain completely separate in the air, they form a purely external mixture. If they collide, coagulate, and condense over time until every single particle contains an identical, uniform blend of soot and sulfate, they form a purely internal mixture.
For decades, global climate models relied on the instantaneous internal mixing assumption to simplify calculations (Liu et al., 2012). Because tracking the exact chemical composition of millions of individual particles is computationally prohibitive, models average the chemical species within a given size range or grid cell. However, treating real-world aerosol populations as binary or instantly uniform introduces massive, systematic biases into climate and air quality predictions (Hughes et al., 2018).
To fix these errors, a growing body of research focuses on accurately representing the aerosol mixing state—the true chemical distribution across individual particles—and tracking how it evolves over time.
The Optical Bias: The Lensing Effect
Why does the internal mixing shortcut fail so dramatically? The first reason is optical.
When a climate model assumes instantaneous internal mixing, it mathematically models an aerosol particle as a concentric sphere: a highly light-absorbing black carbon core surrounded by a shell of reflective, scattering materials like water, sulfate, or nitrate. This shell acts like a magnifying glass. Incoming solar radiation enters the clear shell, refracts, and focuses directly onto the dark core. This phenomenon, known as the lensing effect, artificially boosts the particle's total light absorption (Zhang et al., 2014).
In reality, fresh emissions—such as smoke plumes from wildfires or traffic exhaust—start out largely unmixed or externally mixed (Willis et al., 2016; Brown et al., 2021). Because many light-absorbing cores remain uncoated for days, they absorb far less light than traditional models predict.
By correcting this mixing state bias and representing fresh smoke as an external mixture, Brown et al. (2021) demonstrated that biomass burning aerosols actually exert a net cooling effect on the global climate, correcting a widespread warming bias in state-of-the-art models. Similarly, Feng et al. (2013) showed that accounting for the specific mixing states of organic components like "brown carbon" can fundamentally flip our understanding of an aerosol's net climate impact.
The Microphysical Bias: Cloud Activation and Scavenging
The second failure of the internal mixing assumption relates to cloud microphysics.
Aerosols act as the seeds for clouds; they are the Cloud Condensation Nuclei (CCN) around which water vapor condenses. However, different chemical species interact with water differently. Pure black carbon is hydrophobic (water-repelling), meaning it is a poor cloud seed. Sulfate and sea salt are hydrophilic (water-attracting), making them excellent cloud seeds.
When a model assumes instantaneous internal mixing, it spreads the hydrophilic materials evenly across all particles. This artificially increases the water affinity (hygroscopicity) of otherwise hydrophobic particles (Lee et al., 2016). As a result, models overestimate how easily particles activate into cloud droplets, particularly at high atmospheric supersaturations (Yu et al., 2025).
This overestimation triggers a cascading error known as nucleation scavenging bias. Because models predict that black carbon particles activate into cloud droplets too quickly, they also predict that these particles are prematurely washed out of the atmosphere by rainfall. Ching et al. (2018) discovered that assuming an internal mixture can overestimate the wet removal of black carbon by over 1000% under specific atmospheric conditions. This distorts our calculations of how long warming particles survive in the atmosphere to absorb sunlight.
Quantifying the Continuum via Information Theory
To move beyond simplistic binary classifications, atmospheric scientists needed a way to mathematically define the messy continuum between external and internal mixtures. Nicole Riemer (2013) solved this by adapting diversity measures from ecology and information-theoretic entropy to create the mixing state index, $\chi$.
The metric works by comparing two levels of chemical diversity:
- Average single-particle diversity ($D_\alpha$): The variety of chemical species found within an individual particle, averaged across the population.
- Bulk population diversity ($D_\gamma$): The total variety of chemical species present across the entire aerosol population combined.
The mixing state index is defined as:
$$\chi = \frac{D_\alpha - 1}{D_\gamma - 1}$$
- When $\chi = 0\%$, the population is a purely external mixture (every particle is chemically pure, so single-particle diversity is minimal).
- When $\chi = 100\%$, the population is a purely internal mixture (every particle perfectly mirrors the bulk composition of the entire atmosphere).
This quantitative framework allows researchers to map exactly when and where simplified models break down. For instance, Ching et al. (2017) demonstrated that for highly internal mixtures (high $\chi$), ignoring the mixing state causes negligible errors. However, for externally mixed environments (low $\chi$), neglecting the mixing state causes CCN prediction errors ranging from a 40% underestimation to a 150% overestimation.
Advanced Modeling Paradigms
To capture this chemical continuum, researchers have pioneered two advanced modeling architectures:
1. Source-Oriented Modeling
Instead of grouping particles solely by size, source-oriented models track particles based on their specific origins, such as diesel exhaust, wood smoke, or industrial emissions (Zhang et al., 2014). By maintaining the unique chemical identities of these sources, the models prevent artificial, instantaneous blending. This approach yields highly realistic simulations of atmospheric aging, which significantly improves the accuracy of regional weather, fog, and precipitation forecasts (Lee et al., 2016; Lee et al., 2019).
2. Particle-Resolved Modeling
The most precise approach is particle-resolved modeling, exemplified by frameworks like WRF-PartMC (Curtis et al., 2024). These models use stochastic algorithms to simulate the life cycles of tens of thousands of individual, discrete particles as they undergo coagulation and condensation. While computationally intensive, they provide an exact, uncompromised representation of the mixing state index $\chi$, serving as a vital benchmark for validating simpler models and real-world mass spectrometry observations (Healy et al., 2014).
Scaling Up with Machine Learning
While particle-resolved simulations provide unprecedented accuracy, they are far too computationally heavy to run directly inside global Earth system models, which must simulate decades of global climate data using simplified modal frameworks (Zheng et al., 2021).
To bridge this gap, researchers are turning to machine learning emulators. By running high-fidelity, particle-resolved simulations across a wide range of environmental scenarios, authors have trained machine learning models to predict the evolution of the mixing state index $\chi$ based on standard macro-scale variables (Hughes et al., 2018).
These machine learning models act as a computationally efficient bridge. They allow global climate models to retain their fast, simplified structures while dynamically adjusting for mixing state biases, ultimately leading to far more accurate predictions of global radiative forcing and cloud formation.
Key Foundational Work
Quantifying aerosol mixing state with entropy and diversity measures (2013) — contributed 2 concept elements to this direction.
Brown carbon: a significant atmospheric absorber of solar radiation? (2013) — contributed 1 concept elements to this direction.
Toward a minimal representation of aerosols in climate models: description and evaluation in the Community Atmosphere Model CAM5 (2012) — contributed 1 concept elements to this direction.