Introduction

Maritime navigation plays a pivotal role in global trade, facilitating the movement of goods across interconnected shipping routes1,2,3. The Red Sea, serving as a critical maritime corridor connecting the Mediterranean Sea to the Indian Ocean through the Suez Canal, accommodates approximately 10% of global maritime commerce4. However, this region is increasingly vulnerable to environmental hazards, particularly frequent and intense dust storms5. These storms, originating from the arid expanses of the Arabian Peninsula and driven by seasonal northwesterly Shamal winds, can rapidly reduce visibility to less than 1 km 6,7,8. For example, A major dust storm in April 2015 originated over the Arabian Peninsula, affected multiple countries across the Middle East, and caused widespread visibility drops, transportation disruptions, and flight cancellations9,10. Such extreme reductions in visibility significantly elevate the risks of vessel collisions, groundings, and costly operational disruptions, underscoring the need for improved forecasting and risk management frameworks11.

Globally, dust transport is a major atmospheric process influencing climate, air quality, and visibility12,13. Large dust emissions originate primarily from the Sahara Desert, the Arabian Peninsula, and parts of Central Asia, together accounting for approximately 75% of global dust load14. The Arabian Peninsula, in particular, has been identified as a key source region due to its extensive deserts and prevailing wind systems that drive frequent dust mobilization events15. The Arabian Peninsula stands out as a particularly significant source due to its extensive deserts, limited vegetation cover, and prevailing wind systems that efficiently mobilize and transport dust over long distances16. Despite the ecological and climatic relevance of these emissions, research has disproportionately focused on terrestrial and atmospheric impacts17,18,19, while the direct effects of dust on maritime visibility in the Red Sea remain underexplored20.

Advances in remote sensing technologies have created new opportunities to monitor dust transport dynamics over large and remote ocean areas21,22. Instruments such as the Moderate Resolution Imaging Spectroradiometer (MODIS) provide near-daily aerosol optical depth (AOD) measurements23, while the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) offers vertical aerosol profiles critical for understanding the structure of dust layers24,25. Additionally, the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA-2) supplies high-resolution meteorological reanalysis data essential for capturing wind-driven dust mobilization. However, despite these technological advances, previous studies have struggled to develop robust quantitative models linking aerosol concentrations to visibility reduction over maritime regions22. This limitation primarily arises from technical challenges, including the fusion of heterogeneous datasets with different spatial and temporal resolutions26, limited availability of reliable in situ visibility observations over water21, and the complexity of modeling nonlinear interactions between aerosol microphysical properties and light extinction. As a result, quantitative models explicitly linking AOD and meteorological parameters to visibility degradation remain scarce, especially in maritime environments.

Traditional statistical and physically based modeling approaches for dust transport and visibility prediction often depend on simplifying assumptions about aerosol particle size distributions, optical characteristics, and meteorological conditions27,28. These constraints limit their applicability across diverse environmental settings and contribute to model uncertainty. In contrast, machine learning (ML) techniques offer a promising alternative due to their capacity to automatically extract complex spatiotemporal patterns from large, multi-source datasets29,30,31. Recent studies have demonstrated that CNN-based models can outperform conventional approaches in predicting aerosol concentrations and related air quality indices32,33. Nevertheless, applications of such advanced deep learning (DL) frameworks to maritime visibility forecasting remain limited, particularly for risk assessment in dust-prone maritime corridors such as the Red Sea.

This study aims to address three major research gaps and key innoviation of the study is (i) the absence of integrated multi-source satellite frameworks that combine horizontal AOD measurements and vertical aerosol profiling with meteorological reanalysis data, (ii) the lack of quantitative models that explicitly relate AOD and meteorological variables to maritime visibility degradation, and (iii) the limited application of CNN-based predictive techniques to support risk assessment and navigation safety in dust-prone maritime corridors. This research develops a dynamic methodology to monitor dust transport and assess its impacts on maritime visibility in the Red Sea by utilizing MODIS, CALIPSO, and MERRA-2 data within a DL framework, thereby contributing to safer and more resilient maritime operations.

Study area

The Red Sea region, situated between northeastern Africa and the Arabian Peninsula, is characterized by its unique climatic, geographical, and oceanographic features (Fig. 1). It stretches approximately 2,250 km in length and up to 355 km in width, with an average depth of about 490 m and a maximum depth exceeding 2850 m 34–37. The Red Sea is bordered by countries including Saudi Arabia, Egypt, Sudan, Eritrea, Djibouti and Yemen, and connects the Mediterranean Sea to the Indian Ocean via the Suez Canal and the Bab el-Mandeb Strait.

Fig. 1
figure 1

(source: https://qgis.org).

Geographical overview of the Red Sea region showing major ports and primary shipping lanes. This Figure was created using QGIS 3.12.1 software.

Full size image

The climate of the Red Sea is classified as arid, with extremely high temperatures during the summer months and moderate temperatures in winter. The annual average temperature ranges from 24°C to 30°C. Precipitation is scarce, averaging less than 200 mmyr− 1, and humidity levels vary significantly across seasons. The prevailing meteorological conditions are dominated by seasonal wind patterns that influence both oceanic and atmospheric dynamics. The Shamal winds, in particular, are northwesterly winds that occur frequently during spring and summer, often exceeding speeds of 8–12 ms− 1, contributing to significant dust transport into the Red Sea basin16. Dust storms are frequent and intense in the Red Sea region, particularly during spring and summer (Table 1).

Table 1 Summary of key environmental parameters for the Red Sea region.
Full size table

Studies have indicated that dust concentrations are influenced by the Arabian Peninsula’s vast deserts, such as the Rub’ al Khali, and are transported by the Shamal winds. AOD values during major dust storm events have been observed to exceed 1.0, significantly impairing visibility over the maritime region. In contrast, AOD values during non-dusty periods typically range from 0.1 to 0.3. Wind dynamics are a critical factor in dust mobilization, with wind speeds frequently exceeding thresholds necessary for dust entrainment (> 5 ms− 1) 38. The region’s topography, characterized by coastal mountain ranges and narrow straits, further amplifies dust transport dynamics. Funnel effects caused by these topographic features lead to localized intensification of dust concentrations in central and southern parts of the Red Sea. These dynamics are exacerbated by atmospheric instability, particularly during the transitional periods between seasons.

The Red Sea is of immense strategic and economic importance, serving as a major global maritime trade route. It connects the Mediterranean to the Indian Ocean, facilitating the transportation of goods, energy resources, and raw materials. Key ports such as Jeddah in Saudi Arabia and Port Sudan in Sudan play critical roles in international shipping. Approximately 10% of global maritime trade is estimated to pass through the Red Sea, with the Bab el-Mandeb Strait serving as a critical chokepoint for oil and commodity transportation.

Dust storms in the region disrupt maritime navigation by reducing visibility to less than 1 km during peak events, significantly increasing the risk of collisions and groundings. Port operations are also affected, with delays in cargo handling and vessel turnarounds leading to economic losses. The persistent presence of dust storms necessitates enhanced navigation planning and risk assessment frameworks to ensure safe and efficient maritime operations in this strategically vital corridor38.

Materials and methods

This section outlines the materials and methods used to assess dust transport dynamics and visibility reduction over the Red Sea. The section is organized into 6 subsections. First, Dust transport patterns on the Red Sea presents the datasets used in the study, which include MODIS AOD, CALIPSO, and MERRA-2 data, along with supplementary visibility observations. Back-trajectory and source attribution analysis details the research method, describing the preprocessing steps such as normalization and data fusion. In  Wind speed variability and its role in dust transport, the application of CNN for improving predictive accuracy regarding dust transport and visibility. Validation of remotely sensed AOD and wind speed vs. observed data and the relationship between AOD and wind speed introduces the dust transport modeling process using MODIS AOD and wind data from MERRA-2.  Variability of dust transport intensity on the Red Sea discusses the visibility impact analysis based on the relationship between dust concentration and visibility reduction. Finally,  Visibility impacts and navigation risks describes the Mann-Kendall test, which is used to analyze trends in dust transport and its effects on visibility.

Utilized datasets

In this study, multiple datasets were utilized to analyze aerosol concentrations, vertical dust profiles, meteorological conditions, and visibility over the Red Sea region. The primary data sources included MODIS AOD, CALIPSO lidar measurements, MERRA-2 reanalysis data, and supplementary visibility observations (Table 2).

Table 2 Data sources and their key characteristics.
Full size table

The MODIS instrument aboard NASA’s Terra and Aqua satellites provides AOD measurements, which are indicative of aerosol concentrations in the atmosphere39. MODIS captured data across 36 spectral bands, with spatial resolutions of 250 m (bands 1–2), 500 m (bands 3–7), and 1,000 m (bands 8–36). The MOD04_L2 product was employed for aerosol studies, offering a spatial resolution of 10 km × 10 km at nadir, suitable for large-scale monitoring of dust aerosols. This high temporal frequency, near-daily data allows for effective tracking of dust transport patterns across vast regions40. The CALIPSO satellite, launched on April 28, 2006. CALIOP operated with a two-wavelength laser transmitter, capturing backscatter data to determine the vertical distribution of dust layers41,42. The MERRA-2, offered comprehensive meteorological data, including parameters such as wind speed, atmospheric pressure, and humidity. MERRA-2 provided data at a spatial resolution of 0.5° latitude by 0.625° longitude, with temporal resolutions ranging from hourly to monthly averages43,44. Visibility observations were obtained from ground-based meteorological stations around the Red Sea region (Table 3). It should be noted that all datasets used in this study (MODIS AOD, CALIPSO lidar profiles, MERRA-2 reanalysis data, and visibility observations) cover the period from January 2015 to December 2023, ensuring consistency in temporal coverage across data sources. Also, to ensure data integrity, missing or erroneous values in MODIS, CALIPSO, MERRA-2, and visibility datasets were systematically identified. For MODIS AOD, pixels affected by cloud contamination or with retrieval confidence below the “good” quality flag were excluded. CALIPSO gaps caused by orbit coverage limitations were filled using temporally adjacent profiles (± 1 day) where available; otherwise, such gaps were omitted from analysis. For MERRA-2 and ground-based visibility records, short gaps (< 3 days) were linearly interpolated, while longer gaps were excluded to avoid introducing bias. After cleaning, all datasets were temporally aligned on a daily basis to ensure consistency across sources.

Table 3 Utilized meteorological stations in the Red Sea region.
Full size table

It should be mentioned that KAUST Buoy (Station 23020) is owned and maintained by the King Abdullah University of Science and Technology (KAUST) and is located in the central Red Sea.​ Hurghada and Marsa Alam Meteorological stations are located along the Egyptian coast of the Red Sea and are referenced in studies related to wind resources in the region. Also, in cases where direct visibility data were unavailable, proxies were calculated using established relationships between AOD and visibility. The datasets were preprocessed to handle missing values, and temporal alignment was ensured to facilitate accurate analysis.

This data is critical for modeling the dynamics of dust transport and understanding the meteorological conditions that influence dust movement and concentration. Together, these datasets—MODIS AOD, CALIPSO vertical profiles, and MERRA-2 meteorological data—are combined to create a completer and more accurate model of dust transport and its impact on maritime visibility. This multi-source approach enhances the precision of visibility risk assessments in the Red Sea region by combining aerosol concentration data with vertical distribution and meteorological conditions, which are crucial for accurate predictive modeling of dust storms and their effects on maritime navigation.

Research method

To systematically analyze dust transport dynamics, visibility reduction, and their implications for maritime navigation, a multi-step methodological framework is developed (Fig. 2).

Fig. 2
figure 2

Schematic flowchart of the proposed research method of the current study.

Full size image

As the utilized data was discussed in  Dust transport patterns on the Red Sea, the raw data were preprocessed before analysis. The first step involved normalizing the datasets to ensure consistency across different scales, especially for aerosol concentrations and wind speeds. The normalization was done using a min-max scaling method, which transformed the values into the range of [0, 1] for better comparability and integration (Eq. 1).

$$\:{x}_{n}=\frac{{x}_{i}-{x}_{min}}{{x}_{max}-{x}_{min}},$$
(1)

Where: xn​ is the normalized value. xi​ is the original value. xmin​ is the minimum value of the dataset. xmax​ is the maximum value of the dataset. Unlike some other normalization techniques (e.g., Z-score standardization), min-max normalization does not alter the original distribution of the data. It simply rescales it, maintaining the relative relationship between data points, which can be crucial when dealing with environmental data like aerosol concentrations or wind speed, where the original distribution reflects important underlying patterns. After normalizing the data, missing or erroneous values were removed through a data cleaning procedure to ensure the integrity of the dataset. Subsequently, the cleaned and normalized datasets from MODIS AOD, CALIPSO, and MERRA-2 were fused together. This multi-source data fusion allowed for a more comprehensive understanding of dust transport dynamics and its impacts on visibility.

The data were then analyzed using a combination of traditional modeling and ML techniques. Dust transport was modeled using the MERRA-2 meteorological data with the MODIS AOD and CALIPSO vertical profiles to simulate the horizontal and vertical distribution of dust particles. These models provided insights into the movement and concentration of dust in the region, which was essential for estimating the reduction in visibility caused by dust storms. In addition, advanced ML method, particularly CNN, was employed to enhance the predictive capabilities of the models. CNN was trained on the preprocessed and integrated data to identify complex patterns and relationships between aerosol concentrations, meteorological conditions, and visibility reduction. These models were particularly valuable in improving the accuracy of predictions regarding visibility risks in real-time conditions.

To evaluate the accuracy of the models, predictions were compared against actual visibility observations. The performance of the CNN models was assessed using statistical metrics, which helped quantify the model’s ability to predict visibility accurately. Finally, the model outputs were used to propose shipping routes with minimized risk, based on the visibility predictions. By identifying areas with lower dust concentrations and safer visibility conditions, these routes were recommended for safe maritime navigation. The integrated data from MODIS, CALIPSO, and MERRA-2, combined with CNN, allowed for the development of dynamic models that captured both the spatial and temporal variations of dust transport and their effects on maritime visibility.

CNN application

A CNN was developed to classify dust transport patterns and predict their impact on visibility. CNNs were chosen for this study due to their exceptional capability in analyzing spatial data, particularly for extracting meaningful patterns from AOD satellite imagery. Unlike traditional ML models such as random forest or support vector machines, which typically rely on handcrafted features, CNNs can automatically learn and extract spatial features from high-dimensional remote sensing data. Dust transport exhibits distinct spatial patterns, making CNNs particularly suitable for identifying these patterns and mapping their variations across the Red Sea. The convolutional layers in CNNs allow for efficient feature extraction, reducing computational complexity while preserving key spatial relationships in the AOD data. This makes CNNs an optimal choice for classifying and modeling dust distribution dynamics. The CNN model was extended to capture the temporal dependencies in dust transport by incorporating sequential data over time, which was achieved through a series of convolutional and pooling layers designed to learn both spatial and temporal features. This modification allows the CNN model to handle time-series data by leveraging the powerful capabilities of temporal convolution in DL. Thus, CNNs were applied here to extract spatiotemporal features from both the MODIS AOD data and the MERRA-2 meteorological data.

The model was trained using input features derived from MODIS AOD images and MERRA-2 meteorological data, with visibility observations serving as the target variable (Table 4).

Table 4 Input data, output data, and model objective for CNN-based dust transport and visibility prediction.
Full size table

The dataset was split into training and testing subsets, using an 80 − 20 split, with additional validation techniques applied to ensure the model’s robustness and generalization. The training process involved minimizing the mean squared error (MSE) loss function using the Adam optimizer (Eq. 2) 45.

$$\:MSE=\frac{1}{n}\sum\:_{i=1}^{n}({y}_{i}-{\widehat{y}}_{i}.$$
(2)

Where: n was the total number of observations. yi​ was the actual observed value. \(\:{\widehat{y}}_{i}\)was the predicted value. The goal of the training process is to minimize the MSE, ensuring that the predicted values are as close as possible to the true values.

In CNN, the convolution operation was applied to the input data, where the input data x was passed through filters (kernels) to extract relevant features. The convolution operation is represented as (Eq. 3) 46, 47.

$$\:y=Conv2D\left(x,W\right)+b,$$
(3)

Where: x was the input data (e.g., AOD values from MODIS images). W represented the learned filters (weights). b was the bias term. y was the output feature map, which contains the extracted spatial features. The primary goal of this operation is to extract key spatial patterns, such as dust distribution, from the input data. After the convolution, the ReLU (Rectified Linear Unit) activation function is applied to add non-linearity and help the model capture complex patterns. The ReLU function was defined as (Eq. 4):

$$\:ReLU\left(x\right)=\text{m}\text{a}\text{x}(0,x),$$
(4)

ReLU ensured that only positive values are passed through, preventing the issue of vanishing gradients. Max-Pooling was used to reduce the spatial dimensions (height and width) of the feature maps, which helped reduce computational complexity while retaining important features. Max-pooling was defined as (Eq. 5):

$$\:y=\text{M}\text{a}\text{x}\text{P}\text{o}\text{o}\text{l}\left(x\right),$$
(5)

Where x was the input from the previous layer (typically the convolutional layer), and y was the output after applying the max-pooling operation. After the convolutional and pooling layers, the fully connected layers were used to make the final predictions. These layers were essentially a series of linear transformations. The output of the last pooling layer was flattened and passed through fully connected layers (Eq. 6):

$$\:y=Wx+b,$$
(6)

Where: W was the weight matrix. x is the flattened input feature map. b was the bias term. y was the final output of the fully connected layer, which predicted visibility risk or dust concentration. The model was trained by minimizing the MSE loss function, which measures the difference between predicted and actual values:

To comprehensively evaluate model performance, multiple statistical metrics were employed. Root mean square error (RMSE) was used to measure the standard deviation of residual errors, providing insight into large prediction deviations (Eq. 7) 48. While the coefficient of determination (R²) was utilized to assess the proportion of variance explained by the model (Eq. 8) 49. Additional evaluation metrics, mean absolute error (MAE) and mean absolute percentage error (MAPE), were introduced to enhance the robustness of model assessment (Eqs. 9 and 10)50,51.

$$\:RMSE=\sqrt{\frac{1}{n}\sum_{i=1}^{n}{\left({y}_{i}-\widehat{{y}_{i}}\right)}^{2}},$$
(7)
$$\:{R}^{2}=1-\frac{\sum\:_{i=1}^{n}{\left({y}_{i}-\widehat{{y}_{i}}\right)}^{2}}{\sum\:_{i=1}^{n}{\left({y}_{i}-\stackrel{-}{y}\right)}^{2}},$$
(8)
$$\:MAE=\frac{1}{n}\sum\:_{i=1}^{n}\left|{y}_{i}-\widehat{{y}_{i}}\right|,$$
(9)
$$\:MAPE=\frac{1}{n}\sum\:_{i=1}^{n}\frac{\left|{y}_{i}-\widehat{{y}_{i}}\right|}{{y}_{i}}\times\:100,$$
(10)

where, n is the total number of observations. yi​ represents the actual observed values. \(\:\widehat{{y}_{i}}\:\)represents the predicted values. \(\:\stackrel{-}{y}\) is the mean of the observed values.

It should be mentioned to avoid data leakage and account for temporal dependence, the dataset was split chronologically: the period from January 2015 to December 2021 was used for training (80%), while January 2022 to December 2023 was reserved for testing (20%). This ensures that the model’s predictive performance is evaluated on unseen, future time periods rather than randomly sampled observations. Additionally, data were normalized per year to mitigate seasonal variability before model training.

Dust transport modeling

A dynamic dust transport analysis was conducted to examine observed dust movement across the Red Sea, utilizing AOD data from MODIS and wind vector analysis from MERRA-2. Geographic information system (GIS)-based spatial analysis was employed to map dispersion patterns and assess the spatial extent of dust plumes over time. The model utilized MODIS AOD data to represent aerosol concentrations and MERRA-2 wind vectors to drive the trajectories. Backward and forward trajectories were computed to identify dust source regions and predict transport pathways. GIS-based spatial analysis was conducted using ArcGIS 10.2 to map the dispersion patterns and assess the spatial extent of dust plumes over time.

Visibility impact analysis

To establish a quantitative relationship between dust concentration and visibility reduction, a statistical analysis was performed. The Koschmieder formula, which relates visibility (V) to the atmospheric extinction coefficient (σ), was applied (Eq. 11) 52.

$$\:V=3.912/\sigma .$$
(11)

The extinction coefficient (σ) was estimated using AOD values and atmospheric parameters, following methodologies outlined in previous studies53. This approach allowed for the quantification of visibility reduction attributable to varying dust concentrations.

Mann-Kendall test

The Mann-Kendall test was applied to assess trends in the time-series data related to dust transport and its impact on visibility reduction over the Red Sea region. This non-parametric test was chosen due to its ability to detect monotonic trends in environmental data, which often exhibit non-normal distributions and high variability. In this study, it was used to detect any significant changes in aerosol levels (such as AOD) and visibility in the Red Sea region. The Mann-Kendall test worked by calculating the S statistic, which summed up the signs of the differences between all pairs of data points in the time series, as below (Eq. 12):

$$\:S=\sum\:_{i=1}^{n-1}\sum\:_{j=i+1}^{n}sign({x}_{j}-{x}_{i}),$$
(12)

Where: xi and xj represented the values in the time series. sign was the function that indicates whether xj is greater than, equal to, or less than xi. The input data for the Mann-Kendall test included time-series measurements of MODIS AOD, and meteorological variables like wind speed. These datasets were analyzed to evaluate the changes in aerosol concentrations and visibility over time. The data was taken for multiple years and assessed to determine whether trends in dust transport and its effects on visibility were statistically significant. Once the S statistic was calculated, the Z statistic was used to test the null hypothesis that there was no trend in the data. If Z was greater than a critical value, the null hypothesis was rejected, suggesting that a significant trend exists in the data (either increasing or decreasing).

Back-trajectory and source attribution analysis

Understanding the origin of aerosols is essential for interpreting spatial and temporal variations in AOD and assessing their implications for maritime visibility. Back‑trajectory analysis provides a means to trace the pathways of air parcels contributing to observed aerosol loading, thereby linking measured AOD levels to specific dust source regions54.

This study employed the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model to perform back-trajectory analysis for representative high-AOD episodes identified in each year from 2015 to 2023. Trajectories were initialized at 500 m above ground level for three representative Red Sea locations: northern sector (26° N, 37.5° E), central sector (22° N, 38° E), and southern sector (18° N, 39° E). The 500 m initialization height was selected because it lies within the planetary boundary layer where dust particles are most concentrated during transport events. Lower‑altitude trajectories (e.g., 100–200 m) can be overly sensitive to local turbulence and topographic effects, whereas higher altitudes (> 1 km) often represent free‑tropospheric transport pathways that do not directly affect near‑surface visibility. Thus, 500 m provides a balance between capturing regional dust transport and representing conditions relevant to maritime operations. Each trajectory spanned 72 h prior to the observed AOD peak. This duration was chosen because most Red Sea dust transport episodes are influenced by synoptic‑scale systems (e.g., Shamal winds, cyclonic inflow) that typically evolve within 2–4 days. A 72‑hour window captures the full path from the source regions (e.g., Arabian Peninsula, Nubian Desert) to the Red Sea while avoiding excessive uncertainty associated with longer back‑trajectory calculations. Trajectories were computed using NCEP/NCAR reanalysis meteorological data, ensuring consistent temporal coverage and resolution for all study years. This analysis allowed us to associate high‑AOD events with dominant dust sources and prevailing wind systems, enhancing the interpretation of interannual and spatial variability observed in the MODIS‑derived AOD data.

Results and discussions

Dust transport patterns on the Red Sea

AOD is a fundamental parameter for assessing atmospheric aerosol concentrations, particularly in arid regions where dust storms significantly influence air quality and visibility. AOD quantifies the extent to which airborne particles, such as desert dust, attenuate solar radiation, thereby serving as a reliable proxy for tracking dust transport dynamics. Elevated AOD values typically correspond to intense dust storm events, where high concentrations of suspended particles persist in the atmosphere for extended periods. Conversely, lower AOD values indicate reduced aerosol loading, often corresponding to calmer meteorological conditions or post-dust event sedimentation.

The spatiotemporal distribution of AOD across the Red Sea from 2015 to 2023, as illustrated in Fig. 3, highlights significant interannual variability, reflecting the complex interplay of synoptic-scale meteorology, surface wind patterns, and regional climate variability.

While AOD is a widely used proxy for aerosol loading, it represents the columnar concentration of particles in the atmosphere and does not directly measure the dynamic aspects of dust movement. In contrast, dust transport intensity reflects the combined influence of wind speed, direction, and storm frequency on the mobilization and advection of dust particles. Thus, while high AOD values often coincide with intense dust transport, the two metrics are not identical: AOD provides a measure of aerosol burden, whereas dust transport intensity captures the underlying meteorological processes driving dust emission and movement.

Understanding these variations is critical for maritime safety, as dust-induced visibility reduction poses substantial risks for navigation. Given the economic significance of the Red Sea as a major shipping corridor, real-time monitoring and predictive modeling of dust transport patterns are essential for mitigating potential maritime hazards.

Fig. 3
figure 3

(source: https://qgis.org). Data Source: MODIS Aerosol Optical Depth (MOD04_L2) obtained from Google Earth Engine (GEE) (https://earthengine.google.com/).

Annual spatial distribution of AOD over the Red Sea (2015–2023) The map was processed in QGIS 3.12.1 software.

Full size image

The spatial distribution of AOD values in the Red Sea exhibits distinct interannual fluctuations, with persistent high-AOD regions coinciding with known dust source areas. In 2015, moderate AOD values were observed, particularly in the central and northern Red Sea. The dust activity appeared less intense, with scattered high-AOD patches near coastal regions, likely influenced by localized wind-driven dust mobilization. In 2016, a notable increase in AOD was observed in the southern Red Sea, suggesting a shift in dust transport patterns. This previous studies supported this idea by showing that AOD is a reliable proxy for atmospheric dust loading and storm intensity26,55. This deviation may be attributed to enhanced cyclonic activity and stronger southerly winds, which facilitated dust advection from the African deserts. In 2017, the AOD distribution showed a significant intensification, particularly in the northern and central Red Sea. The presence of widespread high-AOD regions aligns with the intensification of the Shamal winds, which are known to transport dust from the Arabian Peninsula towards the sea. In 2018, a relative decline in AOD values was observed, though certain pockets of high aerosol concentration persisted in the central basin. This decline suggests a weaker dust storm season, potentially due to reduced wind speeds or increased atmospheric stability. In 2019, marked by a resurgence of high-AOD values, particularly in the eastern sector of the Red Sea, indicating strong dust emissions from the Arabian Peninsula. This pattern suggests anomalous wind activity, which may have extended the duration of dust transport events. In 2020, the highest AOD levels of the decade were recorded, particularly in the southern Red Sea. The peak in aerosol concentration correlates with stronger than average surface winds and prolonged dust events, which were documented in regional meteorological records. In 2021, a significant reduction in AOD was observed, consistent with weaker synoptic-scale forcing mechanisms that regulate dust mobilization. This trend underscores the dynamic nature of dust variability, which is modulated by interannual climate oscillations. In 2022, moderate AOD levels were recorded, showing localized dust activity in the central Red Sea. This distribution indicates that dust transport pathways were less extensive compared to previous years. In 2023, a re-emergence of elevated AOD values, especially in the northwestern sector, suggests renewed dust activity. This pattern aligns with climatological expectations, wherein the interplay of seasonal wind systems and synoptic disturbances dictates aerosol loading in the region.

The observed AOD variations from 2015 to 2023 align closely with regional dust storm climatology, where the interaction of large-scale atmospheric circulation patterns and local meteorological factors govern the frequency and intensity of dust transport. Several key mechanisms justify the spatial and temporal patterns observed in Fig. 3. The Shamal winds, predominant in the northern and central Red Sea, were responsible for transporting dust from the Arabian Peninsula, resulting in higher AOD values in these regions during years with intensified wind activity (e.g., 2017, 2019, and 2023). The Haboob events, characterized by strong surface winds associated with convective activity, played a significant role in dust mobilization during peak AOD years, particularly in 2020, when extreme dust loading was observed. The primary dust sources affecting the Red Sea include the Rub’ al Khali Desert (Empty Quarter) in Saudi Arabia and the Nubian Desert in Sudan and Egypt. The transport of dust from these regions is highly sensitive to surface wind intensities and synoptic pressure patterns. Years with lower AOD values (e.g., 2018 and 2021) suggest that either dust emissions from source regions were weaker or atmospheric settling mechanisms were more dominant. The variation in AOD patterns across the years may be influenced by El Niño–Southern Oscillation (ENSO) phases56, which modulate wind speeds and surface pressure systems over the region. Years with higher precipitation anomalies likely contributed to reduced dust activity, as observed in 2018, when a relative decline in AOD was recorded.

Back-trajectory and source attribution analysis

To further elucidate the geographical variations in AOD distribution, back-trajectory analysis for representative high-AOD episodes in each year across the northern (24°–28° N), central (20°–24° N), and southern (16°–20° N) Red Sea was performed. The results, summarized in Table 5, provide insight into the dominant transport pathways and source regions responsible for interannual and spatial AOD variability.

Table 5 Dominant dust sources and transport pathways for high-AOD episodes (2015–2023).
Full size table

Table 5 summarizes the dominant dust sources, wind systems, and quantitative changes in AOD for representative high-AOD episodes between 2015 and 2023. In addition to mean AOD values, it is now tried to report year-to-year percentage changes and latitude/longitude ranges of the affected sectors to better contextualize these variations. During low-AOD years (e.g., 2015, 2018, 2021), back-trajectories were relatively short, originating primarily from localized Arabian Peninsula sources. These events were associated with weaker synoptic forcing, as reflected by mean AOD values between 0.25 and 0.30 and limited spatial extent. By contrast, high‑AOD years (e.g., 2016, 2017, 2019, 2020, 2023) exhibited longer back-trajectories extending into the Nubian Desert, Sahel, and Rub’ al Khali regions, driven by intensified Shamal or southerly flows. These events resulted in AOD increases of 20–43% year-over-year, with the most extreme change occurring in 2020, where mean AOD in the southern Red Sea rose to 0.55 (+ 38% from 2019). For example, the sharp increase in the southern Red Sea in 2016 (0.35) coincided with trajectories from the Nubian Desert under enhanced southerly winds, while the northern Red Sea AOD peak in 2017 (0.42) corresponded to intensified Shamal episodes. The extreme multi‑source dust inflow in 2020 (0.55) reflected cyclonic disturbances enabling contributions from both Arabian and African source regions (Fig. 3).

Wind speed variability and its role in dust transport

Wind speed plays a crucial role in dust mobilization and transport, particularly in arid and semi-arid regions surrounding the Red Sea. The intensity, direction, and persistence of wind directly influence the frequency and magnitude of dust outbreaks, as well as the extent of their transport over maritime areas. Given the Red Sea’s unique climatic and topographical setting, seasonal wind patterns—such as the Shamal winds, monsoonal circulations, and localized sea breezes—significantly contribute to variations in atmospheric dust concentration.

To assess these variations, annual wind speed patterns for the period 2015–2023 have been analyzed (Fig. 4). The spatial distributions reveal distinct interannual trends, highlighting regions of consistent high-wind activity and areas where fluctuations in wind strength align with observed changes in AOD. The interaction between wind speed anomalies and dust transport efficiency is further explored through quantitative analysis of wind variations, considering their implications for aerosol dispersion and maritime visibility.

Fig. 4
figure 4

(source: https://qgis.org). Data Source: MERRA‑2 reanalysis wind speed data obtained from GEE (https://earthengine.google.com/), an open‑access platform.

Spatiotemporal distribution of annual mean wind speeds over the Red Sea (2015–2023). The map was processed in QGIS 3.12.1 software.

Full size image

The wind speed distribution across the Red Sea exhibits significant variability between years, with localized intensifications and periods of attenuation influencing the spatial extent of dust transport.

In 2015, the wind speed was relatively high over the northern and central Red Sea, particularly along the eastern coast. This aligns with the high AOD values observed in Fig. 3, suggesting that strong northwesterly Shamal winds were a dominant dust transport mechanism during this year. In 2016, a notable reduction in wind intensity was observed across most of the study region, particularly in the southern Red Sea. This decrease in wind activity coincided with a significant drop in dust loading, as reflected in the lower AOD values for the same year. In 2017, wind speed rebounded, especially in the central and southern Red Sea, leading to an increase in dust transport. The strengthening of seasonal Shamal winds likely contributed to the resurgence of elevated AOD levels. In 2018, a relatively uniform wind distribution was observed, with moderate wind speeds persisting across the entire Red Sea basin. This pattern was reflected in a more balanced dust transport regime, without significant spikes or declines in AOD. In 2019, a slight reduction in wind speed, particularly in the southern region, was detected. This coincided with a moderate decrease in AOD, suggesting weaker dust entrainment during this period. In 2020, the weakest wind activity of the analyzed period was recorded, particularly in the northern Red Sea. This was directly associated with the lowest annual AOD levels, further confirming the dependency of dust transport on wind intensity. In 2021, wind activity increased again, particularly in the northern and central regions, contributing to a moderate rise in dust concentrations. In 2022, the strongest wind speeds of the entire dataset were observed, particularly in the southern Red Sea. This aligns with the highest AOD levels recorded in Fig. 3, reinforcing the role of wind as a primary driver of dust transport during extreme meteorological conditions. In 2023, a slight reduction in wind speed compared to 2022 was evident, leading to a corresponding decline in AOD values. However, localized wind intensifications in specific regions contributed to sustained dust transport in some areas.

To further contextualize these trends, it is tried to quantify the mean wind speeds by sector and calculated year‑to‑year percentage changes (Table 6).

Table 6 Mean wind speeds and interannual changes across Red Sea sectors (2015–2023).
Full size table

Table 6 shows that wind speeds across the Red Sea experienced substantial interannual fluctuations, with pronounced differences between sectors and years. The highest mean wind speeds were recorded in 2022 (7.8 m.s− 1 in the southern Red Sea, + 24% vs. 2021), coinciding with the highest AOD values observed in Fig. 3. Similarly, 2017 exhibited strong Shamal winds (7.1 m.s− 1), aligning with elevated dust loading in the northern and central sectors. Conversely, 2016 and 2020 recorded the lowest wind speeds (5.2 m.s− 1 and 4.8 m.s− 1, respectively), corresponding to reduced dust entrainment and lower AOD values. These quantitative findings reinforce the connection between wind intensity and dust transport efficiency: years with stronger synoptic winds (e.g., 2017, 2022) experienced more extensive dust mobilization, while years with subdued wind regimes (e.g., 2016, 2020) saw reduced aerosol loading. This highlights the role of Shamal winds in the north and southerly inflows in the south as key drivers of interannual dust variability over the Red Sea.

Validation of remotely sensed AOD and wind speed vs. observed data and the relationship between AOD and wind speed

Validating remotely sensed data is important to ensure the accuracy and reliability of satellite-derived products for atmospheric studies and environmental monitoring. In this study, the validation of AOD and wind speed against observed data from ground-based measurements provides confidence in the use of these datasets for subsequent analysis and model development, such as in CNNs for risk assessments in the Red Sea region (Fig. 5).

Fig. 5
figure 5

Scatter plots of observed vs. remote sensed (a) Wind speed and (b) AOD (2015–2023).

Full size image

Figure 5a shows the correlation between remotely sensed wind speed and observed wind speed, with the regression equation (y = 0.9348x + 0.4726) and an R2 value of 0.73. Figure 5b also illustrates the relationship between remotely sensed AOD and observed AOD, with the regression equation (y = 0.902x + 0.1121) and an R2 value of 0.65. The positive correlation here is also strong, confirming that the acquired satellite AOD values are consistent with ground-based measurements.

Understanding the interaction between wind speed and dust transport is considered significant in assessing AOD variations in the Red Sea region. Wind serves as the primary mechanism for entraining and dispersing dust particles, making it a key meteorological factor influencing AOD levels. To quantify this relationship, the annual mean AOD and wind speed values were normalized to facilitate direct comparison, ensuring that discrepancies arising from unit differences do not obscure the underlying correlation (Fig. 6).

Fig. 6
figure 6

Correlation between normalized mean wind speed and AOD over the Red Sea (2015–2023).

Full size image

The scatter plot presented in Fig. 6 depicts the relationship between normalized AOD and wind speed over the period 2015–2023. A strong positive correlation is observed, indicating that years with higher mean wind speeds generally correspond to increased aerosol concentrations over the Red Sea. This finding aligns with theoretical expectations and previous research, which have established wind-driven dust mobilization as a dominant factor in dust storm dynamics. Despite this overall correlation, certain years exhibit moderate deviations from the regression trend. For instance, in some periods, AOD values remain relatively high despite lower wind speeds, suggesting the influence of additional meteorological and environmental parameters. The availability of loose sediments is a prerequisite for effective dust entrainment. In years following prolonged dry periods, even moderate wind speeds may result in elevated dust transport due to the abundance of unconsolidated particles. Increased atmospheric moisture can suppress dust emission by stabilizing surface particles, reducing their susceptibility to wind-driven uplift. Conversely, prolonged dry conditions can amplify dust activity even in the absence of exceptionally strong winds.

Variability of dust transport intensity on the Red Sea

The temporal evolution of AOD over the Red Sea from 2015 to 2023 was presented in Fig. 3, where spatial patterns indicate notable year-to-year variations in dust transport intensity. The corresponding annual mean AOD values further highlight these fluctuations, providing a quantitative representation of dust loading across the study period (Fig. 7).

Fig. 7
figure 7

Total annual mean AOD values over the Red Sea reflecting the general aerosol loading (2015–2023).

Full size image

The temporal trend presented in Fig. 7 reflects the year-to-year variability in overall aerosol loading across the region. Peaks in AOD, particularly during 2015, 2017, and 2022, suggest periods of elevated aerosol concentration, which may be linked to intensified dust activity, reduced atmospheric dispersion, or stagnation events. Conversely, years such as 2016, 2020, and 2023 exhibit relatively lower mean AOD values, indicating reduced aerosol presence, likely due to weaker wind conditions or more favorable atmospheric cleansing mechanisms. This figure provides an essential baseline for understanding the general aerosol burden in the Red Sea atmosphere, though it encompasses all aerosol types and does not isolate dust-specific contributions. In the following, Fig. 8 presents the annual dust transport intensity index derived from AOD data, emphasizing the component of aerosol loading that is specifically attributable to dust transport processes.

Fig. 8
figure 8

Annual dust transport intensity index derived from AOD (2015–2023).

Full size image

Unlike the general AOD trends shown in Fig. 7, this figure (Fig. 8) focuses on dust-related events that are typically driven by strong surface winds and synoptic-scale weather systems. The intensity index captures both the magnitude and frequency of dust activity over the Red Sea, offering a more targeted perspective on dust storm dynamics. Notably, 2015 and 2022 exhibit high dust transport intensity, aligning with meteorological records of strong Shamal winds and extended dust events. Interestingly, while 2023 has a lower mean AOD, it shows a rebound in dust transport intensity, suggesting that fewer but more severe dust events may have occurred during that year. This distinction underscores the value of differentiating between general aerosol presence and dust-specific transport when assessing visibility impacts and planning maritime navigation strategies. The Mann-Kendall trend analysis applied to the AOD time series did not reveal a statistically significant long-term trend, suggesting that the fluctuations are largely driven by interannual meteorological variability rather than a unidirectional increase or decrease in dust transport intensity (Table 7). However, further investigation incorporating climate projections and land-use changes may provide deeper insights into potential future trends.

Table 7 Mann-Kendall trend analysis of AOD and wind speed over the Red Sea (2015–2023).
Full size table
Fig. 9
figure 9

Seasonal variation of mean AOD across Red Sea subregions: (a) Spring, (b) Summer, (c) Autumn, and (d) Winter.

Full size image

In spring, the highest dust loading is observed in the northern Red Sea, where AOD values exceed 0.9. This can be attributed to pre-summer dust outbreaks, which are commonly triggered by the strengthening of Shamal winds and convective instabilities over arid regions. The central and southern Red Sea regions exhibit considerably lower AOD values during this season. In summer, this period is characterized by widespread dust transport, with all three regions experiencing elevated AOD levels. The central Red Sea exhibits the highest dust concentration (~ 1.4), followed closely by the northern and southern zones. The summer peak is associated with enhanced atmospheric turbulence, prolonged dry conditions, and monsoonal influences, which amplify dust emissions from the Sahara and Arabian deserts. In autumn, AOD values begin to decline, particularly in the central Red Sea, while the southern Red Sea experiences its highest seasonal dust loading (~ 1.0). This suggests that dust transport patterns shift southward during post-summer transitions, likely influenced by shifting pressure gradients and declining Shamal wind intensity. In winter, Dust transport is significantly reduced in the central Red Sea but remains elevated in both the northern and southern regions. The presence of moderate AOD levels (~ 0.7–0.8) suggests that winter dust transport is primarily influenced by low-altitude jet streams and localized dust resuspension rather than large-scale synoptic winds. This analysis reveals distinct spatial heterogeneity in dust transport intensity, with seasonal peaks occurring at different locations within the Red Sea basin. The northern sector is more susceptible to spring and summer dust storms, whereas the southern sector experiences greater autumn and winter dust deposition.

Visibility impacts and navigation risks

To quantify the long term visibility trend over the study period, the Mann-Kendall trend test was applied to the AOD time series data, and the results are summarized in the table below. In addition to significance levels, the Mann-Kendall Z-statistic was reported, which indicates the strength and direction of each trend, and the Theil-Sen slope (β), which quantifies the rate of change per year. The results indicates a statistically significant decreasing trend in visibility, suggesting that the frequency and intensity of dust storms have been increasing over the past decade, leading to more frequent visibility reduction events (Table 8).

Table 8 Mann-Kendall trend analysis for AOD and wind speed trends (2015–2023).
Full size table

The Mann-Kendall test results suggest a positive correlation between wind speed and AOD, indicating that increased wind speeds have been associated with higher dust concentrations. This correlation further supports the hypothesis that stronger winds, which are typically characteristic of dust storm events, play a primary role in exacerbating visibility loss in the region.

The results of this analysis underscore the increasing risks posed by dust-induced visibility loss to maritime navigation in the Red Sea. Reduced visibility can hinder safe navigation, especially during peak dust transport periods, which occur during seasonal changes. The statistical correlation between AOD and wind speed highlights that regions experiencing high wind speeds are also more susceptible to increased dust concentrations, elevating the risk of visibility loss (Fig. 10).

Fig. 10
figure 10

(source: https://qgis.org). Data Sources: MODIS AOD (MOD04_L2), CALIPSO vertical aerosol profiles, and MERRA‑2 meteorological variables accessed via GEE (https://earthengine.google.com/), an open‑access platform.

CNN‑generated risk maps for dust-induced visibility reduction over the Red Sea (2015–2023): Risk categories are defined as: high risk (AOD ≥ 1.0 and/or wind speed ≥ 8 m.s− 1), medium risk (0.5 ≤ AOD < 1.0 and/or 5–8 m.s− 1), and low risk (AOD < 0.5 and wind speed < 5 m.s− 1). The map was processed in QGIS 3.12.1 software.

Full size image

In this study, the risk maps displayed in Fig. 10 were generated using CNN, which was trained to predict visibility reduction over the Red Sea due to dust storms. The CNN model was designed to take AOD and wind speed maps as input features. These two parameters are critical in influencing visibility during dust storms, as AOD represents the concentration of aerosol particles in the atmosphere, and wind speed governs the intensity of dust transport. The CNN model was trained on a combination of these data sources to recognize the spatial patterns of dust movement and its impact on visibility. The model’s output is a set of risk maps, where each pixel represents an area’s predicted visibility level. The risk levels are categorized into high, medium, and low risk zones, with high-risk areas indicating regions with low visibility due to the high concentration of aerosols and stronger wind speeds. To make these categories interpretable, it was defined them using combined AOD and wind speed thresholds derived from model outputs and maritime visibility standards. Specifically, high-risk zones correspond to regions where AOD ≥ 1.0 and/or wind speed ≥ 8 m.s− 1, typically associated with visibility < 2 km. Medium‑risk zones represent areas with AOD between 0.5 and 1.0 and/or wind speeds of 5–8 m.s− 1, corresponding to visibility in the 2–5 km range. Low‑risk zones indicate AOD < 0.5 and wind speeds < 5 m.s− 1, where visibility generally exceeds 5 km (Fig. 10).

Low-risk areas are those with fewer aerosols and weaker winds, resulting in better visibility conditions. By training the model on historical AOD and wind data, the CNN is able to learn the relationship between dust concentration, wind patterns, and visibility reduction. The CNN model utilizes these data to predict visibility degradation across different regions of the Red Sea over the study period from 2015 to 2023. The generated maps are useful for assessing the potential risks to maritime navigation. As dust storms reduce visibility, they pose significant operational challenges to shipping, potentially leading to delays, rerouting, or even accidents. Therefore, these risk maps are crucial for improving decision-making in maritime route planning, as they provide a spatiotemporal representation of visibility impairment that can be predicted in real-time. By comparing these CNN-generated risk maps with observed AOD and wind data, it can be concluded that the CNN model provides a reliable and robust method for predicting visibility reduction in dust-prone areas. The model can be further utilized for risk assessment and maritime safety optimization, allowing for timely interventions when visibility conditions are most critical. As shown in the map for each year, the spatial variation of risk clearly highlights the regions with significant visibility reduction due to dust storms. For example, the high-risk areas, mostly visible in the northern and central Red Sea, correlate with years like 2015, 2020, and 2022, which were marked by more intense dust storms. Conversely, years such as 2016 and 2019, which show lower risk areas, align with reduced dust events, thus providing safer navigation conditions. This demonstrates how the seasonal fluctuations in dust transport contribute to the dynamic nature of maritime risks in the region.

Model performance evaluation and prediction accuracy

In relation to the model performance, the CNN model was designed to predict the variability in visibility degradation caused by dust storms, based on AOD data and wind speed (Fig. 11).

Fig. 11
figure 11

AOD simulation via CNN.

Full size image

The data shown in Fig. 11 further corroborates the model’s performance, as it validates the accuracy of CNN model. The fluctuations in AOD values, as shown in the figure, are indicative of the annual dust storm intensity and its corresponding visibility impact. The model was able to capture this variability, as evidenced by the moderate error values, which confirm that the CNN model was successful in estimating AOD and visibility changes. However, certain years, such as 2023, saw higher prediction errors, which could be attributed to more intense dust storms or changes in atmospheric conditions not fully captured by the model. This highlights areas for potential improvement, including the incorporation of additional atmospheric variables, such as soil moisture and atmospheric turbulence, to improve the model’s predictive capability. Thus, the figure serves to validate the effectiveness of the CNN model in predicting the spatiotemporal variability of AOD, demonstrating its ability to model the dynamics of dust transport and contribute to visibility risk assessments for maritime navigation. In the following, the predictive model developed in this study was evaluated using four key performance metrics: RMSE, R², MAE, and MAPE. These metrics assess how well the CNN-based model predicts visibility reduction due to dust storms over the Red Sea (Table 9).

Table 9 Annual performance metrics of the CNN model for predicting visibility reduction (2015–2023).
Full size table

The results, summarized in Table 9, indicate that RMSE values range from 0.35 to 0.43, suggesting moderate prediction errors in estimating visibility degradation. The R² values between 0.60 and 0.69 demonstrate that the model successfully captures the variability in dust-induced visibility loss, although some external atmospheric factors might still influence the predictions. The MAE values (0.26 to 0.33) show relatively small absolute errors, reinforcing the model’s effectiveness in estimating visibility reductions. Additionally, MAPE values between 7.2% and 8.5% confirm that the model maintains a reasonable level of accuracy across different years.

The slight fluctuations in RMSE and R² indicate that model performance is sensitive to seasonal and interannual variations in dust transport. The best prediction accuracy was achieved in 2018 (RMSE = 0.36, R² = 0.67, MAPE = 7.5%), while the highest prediction errors were recorded in 2023 (RMSE = 0.43, R² = 0.61, MAPE = 8.5%), likely due to increased dust storm intensity and the influence of additional meteorological factors. These results highlight the model’s strong predictive capability while also identifying areas for improvement, such as incorporating additional meteorological inputs (e.g., soil moisture and atmospheric turbulence) to enhance predictive accuracy. The inclusion of MAE and MAPE as evaluation metrics further strengthens the robustness of the model assessment, ensuring a comprehensive analysis of prediction errors.

A key objective of this study was to assess how integrating complementary datasets improves the performance of CNN-based visibility prediction compared to relying on a single satellite product. To quantify this, two CNN models were evaluated. A baseline model using only MODIS AOD as input, and an integrated model combining MODIS AOD with CALIPSO vertical aerosol profiles and MERRA‑2 reanalysis variables. Table 10 summarizes the averaged comparative performance of the two model configurations across the full study period (2015–2023).

Table 10 CNN model performance, Single-source vs. multi-source data integration.
Full size table

The integrated multi-source CNN model outperformed the MODIS-only model across all metrics, achieving a 10% increase in R², a 41% reduction in RMSE, and a 45% and 7% decrease in MAE and MAPE, respectively. These improvements highlight the complementary value of incorporating CALIPSO and MERRA-2 data alongside MODIS. Specifically, MODIS provides broad horizontal coverage of AOD, CALIPSO adds critical information on vertical aerosol layering (which influences near‑surface visibility), and MERRA-2 supplies meteorological context such as wind speed and humidity, which affect dust entrainment, transport, and particle hygroscopic growth. By combining these datasets, the CNN developed a richer spatiotemporal understanding of dust plume dynamics and their impact on visibility, allowing for more accurate predictions and improved delineation of high-risk maritime zones. This integrated approach is especially beneficial in the Red Sea, where dust outbreaks often involve multi-layer transport, strong seasonal wind patterns, and interactions with regional meteorological systems.

The CNN model developed in this study extends analysis by offering practical tools for maritime navigation and operational risk management. The high-resolution visibility risk maps generated by the model can be directly incorporated into decision-support systems for shipping companies and port authorities, enabling the identification of hazardous routes and planning for alternative navigation paths during dust storm events. For example, the ability to predict high-risk zones (AOD ≥ 1.0; wind ≥ 8 m.s− 1) up to several days in advance allows vessels to adjust their transit schedules or select safer corridors, reducing the likelihood of collisions and delays. Additionally, real-time integration of the CNN outputs with Automatic Identification Systems (AIS) and maritime traffic management dashboards could enhance situational awareness for ships operating in the Red Sea. This predictive framework provides a quantitative basis for issuing early warnings to mariners, which is critical in a region where dust-induced visibility loss can drop below 2 km, posing severe navigation hazards.

Implications for climate change and air quality monitoring

The observed AOD variations between 2015 and 2023 indicate that dust storm activity fluctuates in response to meteorological factors, particularly wind speeds and regional atmospheric circulations. The peak AOD years (2015, 2017, 2022) align with periods of intensified Shamal winds, suggesting that increased surface wind stress is a primary driver of dust transport. While the Mann-Kendall trend analysis did not reveal a statistically significant long-term increase in AOD, the extreme AOD event in 2022 aligns with an intensification of regional wind anomalies. Climate projections indicate that rising global temperatures will enhance aridification and wind-driven dust emissions, increasing the probability of extreme dust storms in the future.

The seasonal AOD trends identified in this study (see Fig. 2) provide crucial insights for air quality forecasting in the region. Spring and summer months exhibit the highest dust transport intensity, particularly in the northern and central Red Sea (AOD > 1.0). Air quality monitoring agencies can use this seasonal pattern to issue early warnings for increased PM10 and PM2.5 pollution levels.

The strong correlation (R2 = 0.6) between AOD and wind speed suggests that real-time wind data can serve as a predictive indicator for dust-related air quality degradation. This relationship can be integrated into regional air pollution models to improve dust storm forecasting. Maritime visibility loss due to high AOD levels (> 0.9) presents a significant risk to shipping and port operations. The study’s findings highlight the need for integrating dust storm forecasting into maritime safety protocols.

Limitations and future considerations

While this study provides a comprehensive assessment of dust transport over the Red Sea, several limitations should be acknowledged in terms of data resolution, temporal coverage, and modeling assumptions. These constraints influence the interpretation of numerical results and their practical applications, particularly for air quality monitoring and maritime navigation.

The spatial resolution of satellite datasets used in this study imposes certain limitations on the precision of dust transport assessments. MODIS AOD data has a spatial resolution of 10 km × 10 km, which is sufficient for large-scale dust trend analysis but fails to capture localized dust variability in narrow maritime corridors, such as the Bab el-Mandeb Strait and coastal ports. MERRA-2 reanalysis meteorological data is provided at a 0.5° × 0.625° grid, which limits the ability to resolve fine-scale wind variations affecting dust entrainment. The underrepresentation of small-scale dust emissions and short-lived dust storms may lead to uncertainties in annual and seasonal AOD trends (Figs. 3 and 5). Higher-resolution data from Sentinel-5P (spatial resolution ~ 3.5 km) and ground-based LIDAR systems should be incorporated in future studies to improve dust detection in critical maritime zones.

The availability of satellite data is subject to missing observations due to cloud cover, orbital gaps, and retrieval limitations. MODIS and CALIPSO do not provide continuous daily coverage over all areas, resulting in gaps in AOD measurements, particularly during extreme weather conditions. Cloud cover over the Red Sea obscures aerosol retrievals, leading to potential underestimation of AOD during peak dust events. Certain seasonal AOD peaks (e.g., during major dust storms) may not be fully captured, leading to potential biases in annual dust intensity assessments (Fig. 4). The integration of multiple satellite sensors (e.g., Himawari-8, VIIRS, and SEVIRI) using data fusion techniques can provide more continuous dust monitoring to minimize observational gaps.

Several assumptions were made in the modeling process, which introduce uncertainties in the interpretation of visibility and dust transport trends. While AOD provides a strong indicator of aerosol loading, it does not account for variations in particle size, shape, or chemical composition, which can affect light scattering and visibility differently. The study identifies a strong positive correlation between wind speed and AOD (Fig. 7), yet 40% of AOD variability remains unexplained, suggesting the influence of additional factors such as soil moisture, humidity, and atmospheric turbulence. The assumption that higher AOD always corresponds to lower visibility may not fully capture real-world conditions, where factors such as dust layer altitude and aerosol composition play a critical role. Future research should incorporate ground-based visibility sensors, ceilometers, and in-situ dust samplers to improve the accuracy of dust-induced visibility degradation models.

To enhance the reliability and applicability of dust transport assessments in maritime risk management, future research should focus on higher-resolution dust forecasting models, integrating real-time ground-based observations from aerosol monitoring stations along the Red Sea coastline. The incorporation of LIDAR-based vertical dust profiling can provide additional insights into aerosol layer stratification, improving visibility predictions for maritime operations. ML techniques, particularly hybrid models combining CNNs with numerical weather prediction (NWP) models, could refine short-term dust storm forecasting, allowing for proactive navigation planning. Additionally, future studies should explore the impacts of climate change on dust storm frequency, incorporating long-term climate projections to assess potential shifts in dust source regions and transport pathways. From an operational perspective, the findings of this study underscore the need for integrated maritime navigation systems, where AOD data is coupled with real-time wind speed monitoring and satellite-based visibility estimations. Implementing adaptive route optimization protocols and early warning systems for dust storms can significantly reduce navigation risks and enhance maritime safety in the Red Sea.

Conclusions

This study utilized a multi‑source analytical framework integrating MODIS AOD, CALIPSO vertical aerosol profiles, and MERRA‑2 meteorological reanalysis with advanced GIS‑based spatial modeling and CNNs to comprehensively assess dust transport dynamics over the Red Sea from 2015 to 2023. This integrated approach allowed for the detailed characterization of dust entrainment, transport pathways, and their implications for maritime navigation, bridging the gap between atmospheric observation and operational risk assessment. The main findings revealed:

  1. 1.

    Significant interannual and seasonal variability in dust transport, with peak AOD levels occurring in 2015, 2017, and 2022, coinciding with intensified Shamal winds (predominantly northwesterly).

  2. 2.

    Seasonal analysis demonstrated that spring exhibited the highest dust activity in the northern Red Sea, summer dominated the central basin, and autumn and winter shifts brought increased dust transport to the southern regions.

  3. 3.

    The CNN model, utilizing multi‑source inputs, outperformed single‑source configurations, achieving improved accuracy in predicting visibility reduction, thereby enhancing its value for operational maritime risk management. These findings underscore the necessity of incorporating multi‑dimensional datasets in dust prediction models to support real‑time decision‑making for maritime navigation.

Secondary results indicated that while AOD and wind speed exhibited a strong positive correlation (R² > 0.60), deviations from this relationship highlighted the influence of additional climatic factors, such as synoptic‑scale atmospheric circulations, surface soil conditions, and humidity fluctuations. The Mann‑Kendall analysis revealed a statistically significant decline in visibility over the last decade, emphasizing the growing operational risks posed by frequent dust transport events in one of the world’s most critical shipping corridors.

Despite the strengths of this study, several limitations remain. The analysis relied on satellite and reanalysis data, which may not fully capture localized dust events or provide sufficient temporal granularity for real‑time forecasting. Furthermore, while the CNN framework demonstrated robust predictive capability, its performance could be further enhanced by integrating additional environmental predictors, such as soil moisture, atmospheric stability indices, and synoptic circulation metrics. Future studies should aim to expand the modeling framework to include near‑real‑time data assimilation and to assess the long‑term impacts of climate change on dust storm frequency and intensity in the region.