Weaknesses in dust emission modelling hidden by tuning to dust in the atmosphere

Adrian Chappell, Nicholas P. Webb, Mark Hennen, Charles S. Zender, Philippe Ciais, Kerstin Schepanski, Brandon L. Edwards, Nancy P. Ziegler, Sandra Jones, Yves Balkanski, Daniel Tong, John Leys, Stephan Heidenreich, R. G. Hynes, David Fuchs, Zhenzhong Zeng, Marie Ekström, Matthew Baddock, Jeffrey Lee, Tarek Kandakji2021 · DOI

TL;DR

This paper reveals that traditional dust emission models (TEMs) have significant weaknesses in representing how vegetation affects wind erosion, leading to overestimations of dust emissions, especially in vegetated areas. The authors introduce a novel albedo-based dust emission model (AEM) that better captures the dynamic influence of vegetation, offering a more accurate and less tuned approach for Earth System Models.

What Problem Does This Paper Address?

Dust emission is a critical process influencing global climate, food security, and human health. Vegetation plays a key role by reducing wind speed at the surface and sheltering soil. However, many existing dust emission models (TEMs) simplify this by assuming a bare ground and then applying a vegetation correction. This approach, along with tuning models to match observed atmospheric dust (dust optical depth - DOD), has masked fundamental flaws in how these models represent aerodynamic roughness and vegetation's dynamic impact. This leads to inaccurate dust emission estimates, particularly in vegetated regions, and introduces significant uncertainty into Earth System Models (ESMs). The authors aim to highlight these weaknesses and propose a more robust, albedo-based approach.

Methods

The study compares two dust emission modeling approaches:

  • Traditional Emission Models (TEMs): These models use a fixed aerodynamic roughness length (zo) for the landscape and soil, leading to a fixed drag partition (R). Vegetation's effect is applied as a post-hoc reduction factor (E) based on vegetation cover. The models are often "tuned" to match observed atmospheric dust.
  • Albedo-based Emission Model (AEM): This novel approach uses satellite-derived albedo to represent the aerodynamic sheltering effect of vegetation. It directly estimates the soil wind friction velocity (u*s) based on the concept that vegetation's shadow (1-albedo) is proportional to its sheltering effect. This allows for a dynamic, spatio-temporal representation of aerodynamic roughness without pre-tuning or a separate vegetation cover factor.

Both models were evaluated against satellite-observed dust emission point sources (DPS) and dust optical depth (DOD).

Datasets

The study utilizes the following datasets:

  • Dust Emission Point Sources (DPS): Data from previous studies identifying dust emission locations in New Mexico and Texas (Baddock et al., 2011; Kandakji et al., 2020; Lee et al., 2012). This serves as a primary observational benchmark.
  • MODIS Dust Optical Depth (DOD): Monthly data at 1-degree resolution (MOD08 M3 V6.1) used as a proxy for atmospheric dust.
  • ERA5-Land: Hourly wind speed data at 10m height (~11 km resolution) for the period 2001-2020.
  • MODIS Albedo (MCD43A3): Daily data at 500m resolution, used to derive the soil wind friction velocity in the AEM.
  • MODIS Land Cover: Used to mask land/sea.
  • ISRIC Clay Content: Soil texture data.
  • MODIS NDVI: Used in TEM for bare soil fraction calculation.
  • MODIS Snow Cover (MOD10A1) and ERA5-Land Soil Temperature: Used to mask areas where dust emission is inhibited.

Key Insights and Ideas

  • Tuning Hides Weaknesses: The practice of tuning dust emission models to match atmospheric dust (DOD) has masked fundamental issues in how TEMs represent aerodynamic roughness and vegetation's influence.
  • Vegetation is Dynamic: The effect of vegetation on dust emission is not static; it changes with wind speed and vegetation structure. TEMs, with their fixed roughness and vegetation cover factor, fail to capture this dynamism.
  • Albedo as a Proxy for Roughness: Albedo provides a valuable, dynamically varying proxy for vegetation's aerodynamic sheltering effect, enabling a more realistic representation of soil wind friction velocity.
  • Overestimation in TEMs: TEMs tend to overestimate dust emission frequency, particularly in vegetated areas, because they don't adequately account for the dynamic reduction in wind friction velocity caused by vegetation.
  • DPS as a Better Benchmark: While DOD is a measure of dust in the atmosphere, it's influenced by transport and deposition. DPS data, representing actual emission events, provide a more direct and useful benchmark for evaluating emission models.

Key Results

  • DOD vs. DPS: Dust optical depth (DOD) frequency was found to be two orders of magnitude higher than dust emission point source (DPS) frequency, highlighting DOD's limitations as a direct proxy for emission frequency.
  • Model Overestimation: Both TEM and AEM overestimated dust emission frequency compared to DPS, with RMSE values of 27 days/year for TEM and 20 days/year for AEM.
  • AEM Performance: The AEM showed a stronger correlation (R²=0.43) with DPS frequency than TEM (R²=0.48), and a slightly lower RMSE (20 days/year vs. 27 days/year), indicating improved performance without tuning.
  • TEM's Inadequacy in Vegetated Areas: TEMs produced significantly higher dust emissions in vegetated areas compared to AEM, which correctly reduced emissions due to increased roughness.
  • Spatial Differences: AEM identified localized areas of high dust emission, while TEM predicted larger areas of emission, including regions unlikely to be significant dust sources.

Limitations and Future Work

  • Entrainment Threshold Upscaling: A significant limitation identified is the difficulty in upscaling the grain-scale entrainment threshold (u*ts) to the larger grid box scales used in modeling, which likely contributes to overestimation of dust emission frequency. This is flagged as a priority for future research.
  • Infinite Sediment Supply Assumption: TEMs assume an infinite supply of erodible material, which is often unrealistic. This assumption, coupled with the scale issue of u*ts, is a major contributor to overestimation.
  • Wind Speed Data: While ERA5-Land data is used, the paper acknowledges potential regional overestimation of wind speed, which could still influence dust emission estimates.
  • DPS Data Limitations: The authors note that DPS data may not capture all dust emission sources, especially small, high-frequency events, and may have scale dependencies.
  • Albedo-based Drag Partition Calibration: While the albedo-based approach is promising, the calibration using wind tunnel data might not fully represent the complete range of vegetation conditions.
  • Future Work: The authors emphasize the need to resolve the scale mismatch for u*ts and to investigate the impact of the infinite sediment supply assumption. They also suggest that the albedo-based approach, being driven by readily available prognostic albedo, is well-suited for integration into ESMs to reduce uncertainty in dust emission projections.

Summary updated

Abstract

Dust emissions influence global climate while simultaneously reducing the productive potential and resilience of landscapes to climate stressors, together impacting food security and human health. Vegetation is a major control on dust emission because it extracts momentum from the wind and shelters the soil surface, protecting dry and loose material from erosion by winds. Many of the current dust emission models (TEM) assume that the Earth’s land surface is constantly devoid of vegetation, then adjust the dust emission using a vegetation cover reciprocal, and finally calibrate to dust in the atmosphere. We compare this approach with an albedo-based dust emission model (AEM) which calibrates Earth’s land surface shadow to shelter depending on wind speed, to represent aerodynamic roughness spatio-temporal variation. We also compare these dust emission models with estimates of dust in the atmosphere using dust optical depth frequency (DOD). Using existing datasets of satellite observed dust emission from dust point sources (DPS), we show that during the same period, DOD frequency exceeds DPS frequency by up to two orders of magnitude (RMSEDOD = 67 days). Relative to DPS frequency, both models over-estimated dust emission frequency by up to one order of magnitude (RMSETEM = 6 days; RMSEAEM = 4 days) but showed strong relations with DPS frequency suitable for calibrating models to observed dust emission. Theoretically, the TEM is incomplete in its formulation, which despite the pragmatic adjustment using the vegetation cover reciprocal, causes dust emission to be highly dependent on wind speed and over-estimates large (> 0.1 kg m−2 a−1) dust emission over vast vegetated areas. Consequently, the TEM produces considerable falsely positive change in dust emission, relative to the AEM. Since the main difference between the dust emission models is the treatment of aerodynamic roughness we conclude that its crude representation in the TEM has caused large, previously unknown, uncertainty in Earth System Models (ESMs). Our results indicate that tuning dust emission models to dust in the atmosphere has hidden for more than two decades, these TEM modelling weaknesses and its poor performance. The AEM overcomes these weaknesses and improves performance without tuning. In ESMs the AEM can be driven by available prognostic albedo to represent the fidelity of drag partition physics to reduce uncertainty of aerosol effects on, and responses to, contemporary and future environmental change.
BibTeX
@misc{chappell2021weaknesses,
  author = {Adrian Chappell and Nicholas P. Webb and Mark Hennen and Charles S. Zender and Philippe Ciais and Kerstin Schepanski and Brandon L. Edwards and Nancy P. Ziegler and Sandra Jones and Yves Balkanski and Daniel Tong and John Leys and Stephan Heidenreich and R. G. Hynes and David Fuchs and Zhenzhong Zeng and Marie Ekström and Matthew Baddock and Jeffrey Lee and Tarek Kandakji},
  title = {{Weaknesses in dust emission modelling hidden by tuning to dust in the atmosphere}},
  year = {2021},
  doi = {10.5194/gmd-2021-337},
}

Loading related content...

Figure