Stanford University
A giant plume of Saharan dust across the Atlantic Ocean could dampen the formation of hurricanes over the ocean and influence weather in North America.
But thick dust plumes could also cause landfalling storms to bring heavier rainfall and potentially cause more damage, according to a July 24 study. Scientific progress. The study shows a previously unknown relationship between hurricane rainfall and Saharan dust plumes.
“Surprisingly, the main factor controlling hurricane precipitation is not sea surface temperature or atmospheric humidity as traditionally thought. Instead, it is the dust from the Sahara Desert.
The corresponding author says Wang YuanAssistant Professor of Earth System Science Stanford Duer School of Sustainability.
Previous research found Saharan dust transport may decrease dramatically over the next few decades. Hurricane rainfall likely to increase Due to human-caused climate change.
However, there remains uncertainty about how climate change will affect the Sahara and the outflow of dust. How many We expect future hurricanes to bring more rainfall. Other questions surround the complex relationship between Saharan dust, ocean temperatures, hurricane formation, intensity and precipitation. Filling the gaps is critical to predicting and mitigating the impacts of climate change.
“Hurricanees are one of the most destructive weather phenomena on Earth,” Wang said. Even relatively weak hurricanes can produce heavy rains and flood hundreds of miles inland. “For traditional weather forecasting, especially hurricane forecasting, I don't think dust gets enough attention at this point.”
competition effect
Dust can have a competing effect on tropical cyclones, which are classified as hurricanes in the North Atlantic, central North Pacific, and eastern North Pacific when maximum sustained winds reach 74 mph or higher.
“Dust particles can allow ice clouds to form more efficiently in the hurricane's core, producing more precipitation,” Wang explains. He calls this effect microphysical enhancement. Dust also weakens tropical cyclones by blocking solar radiation and lowering sea surface temperatures around the storm's core.
Wang and colleagues first set out to develop a machine learning model capable of predicting hurricane rainfall, then identified the underlying mathematical and physical relationships.
The researchers used 19 years of meteorological data and hourly satellite precipitation observations to predict rainfall amounts for individual hurricanes.
The results show that a key predictor of rainfall is dust optical depth, a measure of the amount of light passing through the dust plume. They revealed a boomerang-shaped relationship in which rainfall increases with dust optical depth between 0.03 and 0.06 and then decreases sharply. In other words, at high concentrations, dust switches from promoting rainfall to suppressing it.
“Generally, when dust levels are lower, the microphysical enhancement effect is more pronounced. If dust levels are high, it can shield more effectively. [the ocean] “What we call the 'radiation suppression effect' will dominate,” Wang said.
Other authors are affiliated with Western Michigan University, Purdue University, University of Utah, and California Institute of Technology
Magazine
scientific progress
DOI
10.1126/sciadv.adn6106
Article title
Dominant role of Saharan dust on tropical cyclone rainfall in the Atlantic basin
Article publication date
July 24, 2024
From EurekAlert!
Abstract
Tropical cyclone rainfall (TCR) has widespread impacts on coastal communities mainly through inland flooding. The impact of global climate change on TCR is complex and controversial. This study uses the XGBoost machine learning model along with 19 years of meteorological data and hourly satellite precipitation observations to predict the TCR of a single storm. The model identifies dust optical depth (DOD) as a key predictor that can significantly improve performance. The model also revealed a nonlinear and boomerang-shaped relationship between Saharan dust and TCR, with TCR peaking at 0.06 DOD and declining sharply thereafter. This shows a transition from microphysical enhancement to radiation suppression at high dust concentrations. The model also highlights meaningful correlations between TCR and meteorological factors such as sea surface temperature and equivalent potential temperature near the storm core. These findings illustrate the effectiveness of machine learning in predicting TCR and understanding its drivers and physical mechanisms.
introduce
Tropical cyclones (TCs) are extreme weather events that cause catastrophic damage around the world (1, 2). According to global and regional climate models, TC rainfall (TCR) is expected to increase with global warming as temperatures rise and the atmosphere's holding capacity for water vapor increases (3–5). A recent study (6) compared the sea surface temperature (SST)-TCR relationship and found that the climate scale (change ratio between rainfall and temperature increase) under future climate warming (5%/K) is smaller than the Clausius-Clapeyron scale (7%/K) ) and significant scaling in the current climate. In addition, recent satellite observations show that rainfall in the core areas of tropical cyclones shows a decreasing trend, while rainfall in the outer areas of tropical cyclones shows an increasing trend (7, 8). In addition to ocean surface temperature and atmospheric water vapor, other environmental factors also modulate regional changes in TCR, including vertical wind shear (9–11), surface roughness changes (12–14) and atmospheric aerosols (15, 16). How environment and climate influence TCR remains unresolved, especially on multi-year to decadal time scales.
Saharan dust travels across the Atlantic on trade winds and is the dominant aerosol type over the tropical Atlantic in summer and early fall.number 17). It can effectively change the atmospheric radiation flux in the shortwave and longwave bands, and participate in the formation of clouds by acting as cloud condensation nuclei (CCN) and/or ice nuclei (IN) (18). It is reported that Saharan dust tends to inhibit the formation of tropical cyclones by cooling sea surface temperatures, thereby reducing the energy supply of tropical cyclones (19, 20). This phenomenon was particularly evident during the peak of European air pollution in the 1970s and 1980s, when dust emissions in the Sahel were thought to have increased due to prevailing drought conditions. This increase in dust transport coincides with a significant decline in Atlantic hurricane activity (8). Another study (twenty one) highlights the close connection between North Atlantic dust and considerable spatial variation in factors such as zonal wind shear, mid-level humidity and sea surface temperature. However, they found minimal correlation between dust optical depth (DOD) and accumulated cyclone energy in the Atlantic Ocean. As the Saharan's dust-laden air masses move westward, they can introduce dry, stable air into tropical environments. This dry air suppresses the moisture and convection needed for tropical cyclones to form. In addition, dust can block solar radiation from reaching the surface, thereby lowering sea surface temperatures. The impact of dust on TCR may be more complex and multifaceted. Similar to anthropogenic aerosols (e.g. sulfates or hygroscopic organic matter), they provide more CCN to the TC system (twenty two), dust can promote the formation of hydrometeors in cloud towers, enhance the vertical movement of rainbands by increasing latent heat release, and lead to more surface precipitation (twenty three). In short, there is no consensus on the impact of dust effects on TCR, and the relative importance of dust effects compared with other meteorological factors remains uncertain.
Current climate models still do not have sufficient spatial resolution to resolve the complex microphysical processes of clouds and precipitation, particularly how aerosol microphysics affects deep convective clouds. A cloud-resolving numerical model is also used to capture complex air-ocean and aerosol-TC interactions (14, twenty four), running these models on multiyear to decadal climate timescales remains challenging given the computational expense. Therefore, the combination of big data and machine learning (ML) provides a promising alternative for disentangling the complex relationship between environmental forcing and TC activity. Previous studies have shown that machine learning has strong predictive capabilities for tropical cyclone occurrence, intensity, precipitation, and rapid intensification (14, 25–27). Although current machine learning research on TC mainly focuses on enhancing prediction and predictive capabilities, machine learning models also have the potential to reveal complex nonlinear relationships between features and response variables. Recent advances in explainable machine learning further enhance the interpretability of these models. Therefore, in this study we first derive a long-term record of TCR, defined as the mean tropical cyclone rainfall rate within 600 km of each TC location (see Materials and Methods), and then aim to: (i) develop a method that can use environmental (ii) identify the most influential environmental forcing variables in the machine learning model and explore their interactions; (iii) specifically, elucidate the role of Saharan dust in the TCR effect. This will be achieved by comparing various ML models with and without dust variables and interpreting their physical meaning through the lens of ML interpretability techniques.
result
Model performance and dust comprehensive effect
Correlation analysis first showed that the correlation between various environmental factors and TCR was very low (coefficients were usually less than 0.06) (Fig. S1). This suggests that traditional statistical methods such as linear regression may not model TCR well, possibly due to the nonlinear relationship between environmental characteristics and TCR. Therefore, we use a more complex machine learning method, extreme gradient boosting (XGBoost) based on an ensemble of decision trees, to build our TCR model. Two different models were developed, one including only traditional meteorological factors and geographical information, and the other adding DOD as another predictor. The results of five-fold cross-validation show that both DOD and non-DOD models provide good out-of-sample prediction ability (about 0.6 right2, Figure 1, A and B), will not overfit the training data. It is worth noting that the DoD model surpasses its counterpart, which can be seen from the higher right2 and reduced root mean square error (RMSE). The difference in conditional medians further highlights the superiority of the DOD model in most TCR spectra, with significant error reductions observed at both light and heavy TCR extremes (shown in Figure 1C ). Both models tend to underestimate heavy TCR to a greater extent than overestimate light TCR. In terms of spatial distribution, we can also observe a systematic improvement from the non-DOD model to the DOD model (Fig. 1D). On average, the absolute error (AEof the non-DOD model) is approximately four times that of the DOD model. This is emphasized by the more frequent occurrence of AE A ratio exceeding 1 means that the AE of the non-DOD model is always greater than the AE of the DOD model.
Read the rest of the open access article here.
related
