Observations suggest that North African dust absorbs less solar radiation than models estimate

Most atmospheric aerosols cool the climate, but mineral dust, along with black and brown carbon, absorbs solar radiation and can contribute to warming. The magnitude of dust shortwave absorption strongly influences its net direct radiative effect and thus whether dust warms or cools the climate. Dust absorption also affects atmospheric stability, circulations, clouds, precipitation, and processes such as tropical cyclone development. Despite this importance, the amount of shortwave radiation absorbed by dust remains highly uncertain due to poor constraints on microphysical properties—size distribution, shape, and mineralogical composition (refractive index). Models typically overestimate fine dust and underestimate coarse particles, often assume sphericity, and use spatially invariant refractive indices, all of which can bias single-scattering albedo (SSA) and AAOD. Remote-sensing inversions (e.g., AERONET) also require assumptions that can introduce uncertainty in retrieved size distributions and refractive indices. This study aims to constrain North African dust absorption at 550 nm by leveraging in-situ SSA measurements to obtain source-resolved imaginary refractive index and combining these with observational constraints on size distribution, loading, and shape to estimate AAOD and evaluate models and AERONET retrievals.

Prior work shows models underestimate coarse dust and overestimate fine dust relative to in-situ observations, leading to potential biases in radiative effects. Coarse, aspherical dust absorbs more than fine, spherical dust, but models commonly assume sphericity. Dust absorption depends on iron-rich minerals (hematite, goethite) whose spatial variability is large; many models nevertheless apply constant refractive indices. AERONET retrievals, widely used for evaluation, can produce size distributions skewed too fine and refractive indices that differ from in-situ mineralogical analyses. Inversion algorithms are underdetermined and sensitive to shape assumptions, contributing to uncertainties in retrieved absorption properties. Given dust’s large contribution to global aerosol absorption, these uncertainties propagate into climate forcing assessments.

The study constrains North African dust AAOD at 550 nm by combining observational constraints on (1) dust size distribution, (2) dust mass loading, (3) dust particle shape (asphericity), and (4) dust complex refractive index. The region is divided into two source areas (Sahara and Sahel) and source-resolved properties are used. Key steps: 1) Constrain size distribution and loading using DustCOMM, which fuses aircraft in-situ size distribution measurements (capturing coarse dust up to 20 µm), satellite and reanalysis products, and a model ensemble. 2) Constrain dust shape using measurement-based distributions of aspect ratio and height-to-width ratio to account for asphericity. 3) Constrain imaginary refractive index at 550 nm by optimizing agreement between estimated dust SSA and a compilation of 14 in-situ SSA measurements over North Africa. Estimated SSA is computed with DustCOMM size distributions, laboratory-based real refractive index n=1.51±0.03, and asphericity, and is collocated in diameter range, altitude, location, and season with each measurement. A cost function minimizing squared differences yields source-resolved imaginary refractive indices. Single-particle optical properties incorporating asphericity are taken from a precomputed database. 4) Compute size-resolved AAOD by integrating the product of size-resolved mass absorption efficiency (dependent on real/imaginary refractive index and shape) and the source-resolved column-integrated dust mass size distribution and loading, normalized to exclude particles >20 µm. 5) Uncertainties are quantified via non-parametric bootstrap sampling that propagates uncertainties from in-situ measurements, observational datasets (DustCOMM), and model spread. Comparisons: Two model ensembles are evaluated: six selected global aerosol/climate models (e.g., GISS ModelE, WRF-Chem, CESM, GEOS-Chem, CNRM/ARPEGE-Climate, IMPACT) and eight AeroCom Phase III models. For comparability, simulated SSA and AAOD are computed over the same diameter, altitude, locations, and seasons as the measurements. Bias decomposition replaces, in turn, model inputs with constrained values (size distribution, loading, shape; then refractive indices) to attribute AAOD biases. AERONET: Version 3 (levels 2.0 and 1.5 where needed) retrievals of size distribution, complex refractive index, and AAOD are screened for dust dominance using Angstrom exponent ≤0.2 (440/870 nm), dust fraction >60% of total extinction (from MERRA-2), sea-salt surface contribution <20%, and sufficient sampling. Imaginary refractive index and AAOD are interpolated to 550 nm via second-order polynomial fits in log–log space. Total AAOD comparisons add AeroCom non-dust AAOD to the constrained dust AAOD for consistency.

• In-situ SSA vs models: Measured SSA at 550 nm over North Africa averages ~0.97 (range 0.92–0.99), whereas selected models and AeroCom models yield ~0.95 (0.94–0.97) and ~0.94 (0.93–0.96), respectively, implying models underestimate SSA by about 5% with RMSE up to a factor of two larger than constrained SSA. - Constrained imaginary refractive index: Mean k(550 nm) for North African dust is 0.0012 (one-standard-error range 0.0009–0.0016), substantially smaller than model assumptions: selected models mean ~0.0029 (0.0014–0.0030) and AeroCom mean ~0.0026 (0.0011–0.0031). AERONET dust-dominated retrievals give ~0.0019 (0.0016–0.0021). Thus, models overestimate k by >2× on average; AERONET is ~54% larger than the constraint. - Constrained dust AAOD (550 nm): Over North African continent mean AAOD ~0.0094 (0.0073–0.0120); selected models ~0.0110 (0.0064–0.0494); AeroCom ~0.0180 (0.0148–0.0219). Over the broader domain where North African dust dominates (>80% of dust load): constrained AAOD ~0.0045 (0.0035–0.0057); selected models ~0.0053 (0.0030–0.0232); AeroCom ~0.0090 (0.0059–0.0107). Hence models overestimate dust absorption by up to a factor of two regionally. - AERONET total AAOD at dust-dominated sites: AERONET average 0.029 (0.021–0.031) vs this study 0.017 (0.010–0.027); overestimation can reach up to ~3× at some stations; differences are ~55% larger at Saharan than Sahelian sites. - Bias decomposition: For selected models, fine dust (D≤5 µm) AAOD is overestimated by ~0.0021, while coarse dust (D≥5 µm) AAOD is underestimated by ~−0.0013 due to size distribution biases (too much fine, too little coarse). Overestimated imaginary refractive index contributes an additional AAOD overestimate of ~0.0023 (0.0006–0.0041). About half of the k-driven overestimate is masked by the coarse-dust underestimation. AERONET retrievals similarly overestimate fine and underestimate coarse modes and have higher k, explaining higher retrieved AAOD.

The study directly addresses the uncertainty in dust absorption by constraining the imaginary refractive index using in-situ SSA and combining it with observationally constrained size distribution, loading, and asphericity. Findings show that models and AERONET retrievals generally overestimate mid-visible absorption by North African dust because they assume higher imaginary refractive indices and have size distributions biased toward fine particles while missing coarse particles. This overestimation inflates simulated and retrieved AAOD and can bias assessments of dust radiative effects. The interplay between parameters partially masks errors: elevated k increases absorption, but underestimated coarse dust decreases it, so net model bias is smaller than it would be from k alone. The results imply that many prior model-based and retrieval-based estimates of dust impacts on energy balance, clouds, precipitation, and biogeochemistry may be biased, underscoring the need to revise dust optical property representations and retrieval assumptions. They also suggest that using AERONET as a benchmark without accounting for these dust-condition biases can propagate errors into satellite algorithm tuning and model evaluation.

Climate and chemical transport models, as well as dust-dominated AERONET retrievals, tend to overestimate mid-visible absorption by North African dust. Constraining the imaginary refractive index to k≈0.0012 at 550 nm (vs ~0.0026–0.0029 in models) and using observed size distributions and asphericity yields lower AAOD estimates than models and AERONET, by as much as a factor of two regionally. Model overestimation of k is partly canceled by concurrent underestimation of coarse dust abundance, masking the full magnitude of the bias. Because North Africa supplies roughly half of global dust, these biases have substantial implications for simulated dust radiative effects, hydrological impacts, and iron deposition. Future work should: improve in-situ observations that capture coarse and super-coarse dust; refine model representations of dust size distributions, mineralogy-dependent refractive indices, and particle shape; reassess AERONET retrieval assumptions for dusty conditions; better constrain non-dust absorbing aerosol burdens when comparing to AERONET; and evaluate spectral and vertical variations of dust absorption across regions beyond North Africa.

Key limitations include: (1) In-situ SSA measurements often miss coarse and super-coarse particles due to instrument cutoffs (e.g., nephelometers, PSAPs), and fine-mode SSA can be contaminated by non-dust aerosols (e.g., black carbon). (2) Input datasets carry uncertainties: DustCOMM size distributions exclude D≥20 µm and rely on seasonal, location-representative constraints that may not match specific measurement dates; uncertainties in dust loading, source apportionment, and shape distributions also propagate. (3) AERONET comparisons use non-dust AAOD from models that may underestimate black carbon, likely reducing apparent discrepancies; in-situ SSA is not perfectly collocated or column-integrated like AERONET. (4) Potential contributions from non–North African sources (e.g., Middle East, Central Asia) are small but non-negligible and omitted; effects of atmospheric aging on refractive index were not explicitly modeled but are suggested to be small by prior studies. Overall, reported uncertainties from bootstrap analyses are lower bounds.