Saharan dust regulates hurricane rainfall – Watts Up With That?

Stanford University

Giant plumes of Sahara Desert dust that gust across the Atlantic can suppress hurricane formation over the ocean and affect weather in North America. 

But thick dust plumes can also lead to heavier rainfall – and potentially more destruction – from landfalling storms, according to a July 24 study in Science Advances. The research shows a previously unknown relationship between hurricane rainfall and Saharan dust plumes. 

“Surprisingly, the leading factor controlling hurricane precipitation is not, as traditionally thought, sea surface temperature or humidity in the atmosphere. Instead, it’s Sahara dust,”

said the corresponding author Yuan Wang, an assistant professor of Earth system science at the Stanford Doerr School of Sustainability.

Previous studies have found that Saharan dust transport may decline dramatically in the coming decades and hurricane rainfall will likely increase due to human-caused climate change. 

However, uncertainty remains around the questions of how climate change will affect outflows of dust from the Sahara and how much more rainfall we should expect from future hurricanes. Additional questions surround the complex relationships among Saharan dust, ocean temperatures, and hurricane formation, intensity, and precipitation. Filling in the gaps will be critical to anticipating and mitigating the impacts of climate change.

“Hurricanes are among the most destructive weather phenomena on Earth,” said Wang. Even relatively weak hurricanes can produce heavy rains and flooding hundreds of miles inland. “For conventional weather predictions, especially hurricane predictions, I don’t think dust has received sufficient attention to this point.”

Competing effects

Dust can have competing effects on tropical cyclones, which are classified as hurricanes in the North Atlantic, central North Pacific, and eastern North Pacific when maximum sustained wind speeds reach 74 miles per hour or higher.

“A dust particle can make ice clouds form more efficiently in the core of the hurricane, which can produce more precipitation,” Wang explained, referring to this effect as microphysical enhancement. Dust can also block solar radiation and cool sea surface temperatures around a storm’s core, which weakens the tropical cyclone.

Wang and colleagues set out to first develop a machine learning model capable of predicting hurricane rainfall, and then identify the underlying mathematical and physical relationships.

The researchers used 19 years of meteorological data and hourly satellite precipitation observations to predict rainfall from individual hurricanes. 

The results show a key predictor of rainfall is dust optical depth, a measure of how much light filters through a dusty plume. They revealed a boomerang-shaped relationship in which rainfall increases with dust optical depths between 0.03 and 0.06, and sharply decreases thereafter. In other words, at high concentrations, dust shifts from boosting to suppressing rainfall.  

“Normally, when dust loading is low, the microphysical enhancement effect is more pronounced. If dust loading is high, it can more efficiently shield [the ocean] surface from sunlight, and what we call the ‘radiative suppression effect’ will be dominant,” Wang said.

Additional authors are affiliated with Western Michigan University, Purdue University, University of Utah, and California Institute of Technology

Journal

Science Advances

DOI

10.1126/sciadv.adn6106

Article Title

Leading Role of Saharan Dust on Tropical Cyclone Rainfall in the Atlantic Basin

Article Publication Date

24-Jul-2024

From EurekAlert!

Abstract

Tropical cyclone rainfall (TCR) extensively affects coastal communities, primarily through inland flooding. The impact of global climate changes on TCR is complex and debatable. This study uses an XGBoost machine learning model with 19-year meteorological data and hourly satellite precipitation observations to predict TCR for individual storms. The model identifies dust optical depth (DOD) as a key predictor that enhances performance evidently. The model also uncovers a nonlinear and boomerang-shape relationship between Saharan dust and TCR, with a TCR peak at 0.06 DOD and a sharp decrease thereafter. This indicates a shift from microphysical enhancement to radiative suppression at high dust concentrations. The model also highlights meaningful correlations between TCR and meteorological factors like sea surface temperature and equivalent potential temperature near storm cores. These findings illustrate the effectiveness of machine learning in predicting TCR and understanding its driving factors and physical mechanisms.

INTRODUCTION

Tropical cyclones (TCs) are extreme weather events that have caused catastrophic damages globally (1, 2). According to global and regional climate models, TC rainfall (TCR) is expected to increase with global warming, following the increased water vapor holding capacity in the atmosphere with rising temperature (3–5). A recent study (6) compared the sea surface temperature (SST)–TCR relationships and discovered that the climate scaling (changing ratio between rainfall and rising temperature) under future warmer climate (5% per K) is smaller than the Clausius-Clapeyron scaling (7% per K) and apparent scaling under current climate. In addition, recent satellite observations revealed a decreasing trend of rain rate in the core part of TCs but increasing trend in outer bands (7, 8). Besides ocean surface temperature and water vapor in the atmosphere, other environmental factors regulate the regional variations of TCR, including vertical wind shear (9–11), surface roughness change (12–14), and atmospheric aerosols (15, 16). How the environment and climate influence the TCR remains unresolved, especially over multiyear to decadal time scales.

Saharan dust, transported across the Atlantic Ocean by trade winds, is the predominant aerosol type during summer and early fall over the tropical Atlantic (17). It can efficiently alter atmospheric radiative fluxes in both shortwave and longwave bands and participate in cloud formation by serving as cloud condensation nuclei (CCN) and/or ice nuclei (IN) (18). It has been reported that Saharan dust tends to suppress the formation of tropical cyclones via a cooling effect on SST that consequently cuts the energy supply for TCs (19, 20). This phenomenon was evident during the peak of European air pollution in the 1970s and 1980s, which is believed to have amplified the Sahel dust emissions due to the prevalent drought conditions. This intensified dust transport coincided with a noticeable downturn in Atlantic hurricane activity (8). Another study (21) has highlighted a close association between the North Atlantic’s dust and considerable spatial shifts in factors such as zonal wind shear, midlevel moisture, and SST. However, they found a minimal correlation between dust optical depth (DOD) and the Atlantic’s accumulated cyclone energy. As Saharan dust-laden air masses move westward, they can introduce dry and stable air into the tropical environment. This dry air inhibits the moisture and convection required for tropical cyclone formation. Moreover, by blocking solar radiation from reaching the surface, dust can reduce SST. The dust effects on TCR can be more complicated and multifaceted. Similar to the anthropogenic aerosols (e.g., sulfate or hygroscopic organics) that provide more CCN to TC systems (22), dust can foster the hydrometeor formation in the cloud tower, enhance the vertical motion of rain bands via elevated latent heat release, and result in more surface precipitation (23). In a nutshell, there is no consensus on the sign of the dust effect on TCR, and it remains uncertain what is the relative importance of dust effect compared to the other meteorological factors.

Current climate models still do not have sufficient spatial resolution to resolve the complex microphysical processes of cloud and precipitation, particularly how aerosol microphysics affects deep convective clouds. While cloud-resolving numerical models were adopted to capture the complex air-sea and aerosol-TC interactions (14, 24), it remains challenging to run these models over multiyear to decadal climate time scales, given their computational expense. Therefore, a combination of big data and machine learning (ML) offers a promising alternative method for untangling those complex relationships between environment forcings and TC activities. Previous studies have demonstrated that ML has robust predictive capabilities in TC genesis, intensity, precipitation, and rapid intensification (14, 25–27). While current ML research on TCs primarily centers on enhancing forecasting and prediction capabilities, ML models also have the potential to unveil intricate and nonlinear relationships between features and response variables. Recent advancements in interpretable ML further bolster the interpretability of these models. Therefore, in this research, we first derive a long-term record of TCR, which is defined as the average tropical cyclone rain rate within 600 km of each TC position (see Materials and Methods), and then aim to: (i) develop an ML model capable of predicting TCR variabilities across the Atlantic Ocean using environmental forcing variables; (ii) pinpoint the most influential environmental forcing variables within the ML model and explore their interactions; and (iii) specifically, elucidate the role of Saharan dust in TCR. This will be achieved by contrasting various ML models with and without the dust variable and interpreting their physical significance through the lens of ML interpretability techniques.

RESULTS

Model performance and overall effect of dust

A correlation analysis first shows very low correlations (coefficient generally smaller than 0.06) between individual environmental factors and TCR (fig. S1). It indicates that conventional statistical methods such as linear regression may not work well to model TCR, likely due to the nonlinear relationships between environmental features and TCR. Therefore, we use a more sophisticated ML approach, the Extreme Gradient Boosting (XGBoost) based on an ensemble of decision trees, to build our TCR models. Two distinct models were developed, one only including the traditional meteorological factors and geoinformation and the other adding DOD as another predictor. Results from fivefold cross-validation demonstrate that both DOD and non-DOD models offer decent out-of-sample prediction capabilities (with ~0.6 R², as in Fig. 1, A and B), without overfitting the training data. Notably, the DOD model surpasses its counterpart, as evidenced by a higher R² and a reduced root mean squared error (RMSE). The differences in conditional median values further highlight the DOD model’s superiority across most TCR spectrums, with pronounced error reduction observed for both light and heavy TCR extremes (as illustrated in Fig. 1C). Both models tend to have a larger magnitude of underestimates for heavy TCR than the overestimates for light TCR. With respect to the spatial distribution, we can also observe systematic improvement from the non-DOD to DOD models (Fig. 1D). On average, the absolute error (AE) of the non-DOD model is approximately fourfold that of the DOD model. This is underscored by the more frequent appearance of an AE ratio exceeding 1, indicating that the non-DOD model’s AE is consistently larger than that of the DOD model.

Read the rest of the open access article here.