Satellite Sensors Research Articles

Transformer-based adaptive 3D residual CNN with sparse representation for PAN-sharpening of multispectral images

ABSTRACT Panchromatic (PAN)-sharpening is an important process for most of the practical appliances in the domain of geospatial analysis. Because of the balance among instantaneous field of view and Signal-to-Noise Ratio (SNR), the Multi-Spectral (MS) and PAN images with complementary features are captured by the satellite sensors. Fusing PAN illustrations with maximal dimensional resolution and MS illustrations with maximal spectral resolution is known as PAN-sharpening, which is widely utilized to improve spatial and spectral resolutions. The crucial point in the PAN-sharpening task is to maintain a balance between the extraction of data and the insertion of data. If a misbalance occurs, it may lead to a distortion of the intensity. To address these issues, a new PAN-sharpening model is developed to combine the MS and PAN illustrations to increase the image’s spatial resolution. In addition, the spatial and spectral distortions of the illustrations are highly decreased by using this PAN-sharpening model. Originally, the PAN and MS illustrations were collected from traditional online sources. The MS and PAN images are initially combined and given to the sparse representation process. This provides a simulated PAN-sharpened image. This image is further provided to the intelligent deep learning network, which is the Transformer-based Adaptive 3D Residual Convolutional Neural Network (TA-3DRCNN) and hence the higher-resolution images are obtained, and it is useful in various applications. The parameters present in TA-3DRCNN are tuned using the Modified Uniform Number-based Gannet Optimization Algorithm (MUN-GOA) for upgrading the effectiveness of PAN-sharpening. The resultant outcome is contrasted with the customary PAN-sharpening approaches to validate the system’s effectiveness with respect to various measures.

Leveraging Remote Sensing Products to Estimate Grassland Productivity in the Canadian Prairies

Abstract Remote sensing is the acquisition of data about an area without being in direct contact with it. It can be used to estimate vegetation productivity, offering a cost-effective and efficient alternative to traditional field measurements. This technology enables timely and consistent monitoring of large areas over time using satellite sensors. More recently, drone-mounted sensors have been used for monitoring smaller areas with higher spatial resolutions. Indices derived from remote sensing datasets can be used to estimate vegetation productivity and health, as well as to monitor grassland ecosystems promptly. By providing detailed and frequent data, remote sensing supports the development of sustainable practices and policies that promote efficient land management, ensuring the long-term health and productivity of grassland ecosystems. A case study of Manitoba’s rangeland ecoregions is provided to illustrate how remote sensing products can be used to quantify grassland distribution in different geographic regions of the province. Information © The Authors 2024

Assessing the impact of dams and unsustainable agricultural development on Lake Urmia drying up using multi satellite sensor and deep learning

Assessing the impact of dams and unsustainable agricultural development on Lake Urmia drying up using multi satellite sensor and deep learning

Detection of Marine Oil Spill from PlanetScope Images Using CNN and Transformer Models

The contamination of marine ecosystems by oil spills poses a significant threat to the marine environment, necessitating the prompt and effective implementation of measures to mitigate the associated damage. Satellites offer a spatial and temporal advantage over aircraft and unmanned aerial vehicles (UAVs) in oil spill detection due to their wide-area monitoring capabilities. While oil spill detection has traditionally relied on synthetic aperture radar (SAR) images, the combined use of optical satellite sensors alongside SAR can significantly enhance monitoring capabilities, providing improved spatial and temporal coverage. The advent of deep learning methodologies, particularly convolutional neural networks (CNNs) and Transformer models, has generated considerable interest in their potential for oil spill detection. In this study, we conducted a comprehensive and objective comparison to evaluate the suitability of CNN and Transformer models for marine oil spill detection. High-resolution optical satellite images were used to optimize DeepLabV3+, a widely utilized CNN model; Swin-UPerNet, a representative Transformer model; and Mask2Former, which employs a Transformer-based architecture for both encoding and decoding. The results of cross-validation demonstrate a mean Intersection over Union (mIoU) of 0.740, 0.840 and 0.804 for all the models, respectively, indicating their potential for detecting oil spills in the ocean. Additionally, we performed a histogram analysis on the predicted oil spill pixels, which allowed us to classify the types of oil. These findings highlight the considerable promise of the Swin Transformer models for oil spill detection in the context of future marine disaster monitoring.

UTHP spectrometer: an ultra-thin and high-performance imaging spectrometer for ocean color remote sensing in broadband

The principle of ocean color detection involves using satellite sensors to capture variations in received signals, which are subsequently analyzed to infer the concentrations of various components responsible for ocean color changes. In recent years, significant research advancements have been achieved, centered on the development of ocean color detectors and the analysis of the resulting data. We have developed an innovative ultra-thin high-performance (UTHP) scanning imaging spectrometer for ocean color remote sensing across the near ultraviolet-visible-near infrared (NUV-VIS-NIR) spectrum. This spectrometer overcomes the bulkiness of conventional equipment by integrating a waveguide platform that replaces traditional optical elements with a single flat glass panel, featuring curved surfaces and high-reflectivity coatings to perform essential optical functions. We believe this to be a novel design that greatly reduces the size and weight of the device, making it highly suitable for deployment on various platforms, including aircraft and satellites. The optical system underwent extensive testing, including simulations using Zemax software and physical prototyping, achieving a spectral resolution of less than 3 nm across the 400-1000 nm waveband while maintaining low distortion. The UTHP spectrometer’s compactness, lightweight design, and exceptional performance mark a significant advancement in imaging spectrometry, with the potential to transform remote sensing by providing more efficient, accessible methods for studying expansive oceanic regions.

AI-based Validation of Deforestation Using High-Resolution Satellite Imagery in the Brazilian Amazon

Abstract. Forests play an important role in the Earth systems for carbon sequestration and climate change mitigation, yet they have been increasingly disturbed by deforestation and forest degradation at an unprecedented pace. The Brazilian Amazon, for instance, experienced a 140% rise in deforestation from 2012 to 2020, with a record loss of 13,200 km2 between August 2020 and July 2021. Alarmingly, 87% of 2019 deforestation alerts occurred on private properties, with 61% in legally restricted areas. Existing deforestation monitoring systems, such as PRODES and the Global Forest Change dataset, use about 30m resolution satellite imagery, which is insufficient for operational validation at fine scales. The Deforestation Alert System by Imazon leverages high-resolution PlanetScope data (3-4m) but faces challenges due to fewer spectral bands and variations in reflectance values across different satellite sensors and dates. As a result, current validation is based mainly on manual inspection which is highly time-consuming. To address the challenges and automate the validation process, this work develops a system based on deep learning – known for its ability to capture complex texture patterns in high-resolution images – to inspect and confirm new deforestation sites. Specifically, the system takes inputs from potential deforestation sites suggested by coarse-resolution products and uses a pair of PlanetScope images before and after the change at each site to determine new deforestation (excluding existing deforestation). Our results demonstrate that the new system achieves robust and high-quality accuracy under various test conditions.

Utilizing Data Mining Techniques in Geophysical and Biological Analysis for Assessing Environmental Risks

This study aimed to assess environmental risks using data extraction techniques. It focused on geophysical and biological factors and addressed the urgent need for effective risk management strategies to reduce soil erosion, water pollution, and air quality deterioration. A comprehensive dataset was created through the systematic collection of geophysical and biological data including temperature, soil composition, and biological abundance index. It used equipment such as satellite sensors and mountain transmitting stations. Various statistical tools used include decision trees and random forest algorithms. They were used to analyze data and identify important environmental risk factors. The results showed some interesting insights, revealing that the Neural Network has an accuracy of 95.5%, and the Random Forest algorithm predicts risk with an accuracy of 92.0%. It analyzed the classification of environmental hazard zones and identified high-risk zones, such as Zone A, which contains 10,000 people affected by erosion and Zone D, 20,000 people who were at risk from soil contamination. The study concludes that social media can significantly improve understanding and management of environmental risks. Additionally, it provides a useful framework for decision-makers and stakeholders to promote sustainable environmental practices.

Identification of Global Extended Pseudo Invariant Calibration Sites (EPICS) and Their Validation Using Radiometric Calibration Network (RadCalNet)

This study introduces a global land cover clustering using an unsupervised algorithm, incorporating the novel step of filtering data to retain only temporally stable pixels before applying K-means clustering. Unlike previous approaches that did not assess the pixel-level temporal stability, this method provides more reliable clustering results. The K-means identified 160 distinct clusters, with Cluster 13 Global Temporally Stable (Cluster 13-GTS) showing significant improvements in temporal stability. Compared to Cluster 13 Global (Cluster 13-G) from earlier research, Cluster 13-GTS reduced the coefficient of variation by up to 1% and increased the number of calibration locations from 23 to over 50. This study also validated these clusters using TOA reflectance from ground-truth measurements collected at the Radiometric Calibration Network (RadCalNet) Gobabeb (RCN-GONA) site, incorporating data from Landsat 8, Landsat 9, Sentinel-2A, and Sentinel-2B. The GONA Extended Pseudo Invariant Calibration Sites (EPICS) GONA-EPICS cluster used for the validation provided statistically comparable mean TOA reflectance to RCN-GONA, with a reduced chi-square test indicating minimal differences within the cluster’s uncertainty range. Notably, the difference in reflectance between RCN-GONA and GONA-EPICS was less than 0.023 units across all the bands. Although GONA-EPICS exhibited slightly higher uncertainty (6.4% to 10.3%) compared to RCN-GONA site (<5%), it offered advantages such as 80 potential calibration points per Landsat cycle and reduced temporal instability, and it provided alternatives to reduce the reliance on single sites like traditional PICS or RCN-GONA, making it a valuable tool for calibration efforts. These findings highlight the potential of the newly developed EPICS for radiometric calibration and stability monitoring of optical satellite sensors. Distributed across diverse regions, these global targets increase the number of calibration points available for any sensor in any orbital cycle, reducing the reliance on traditional PICS and offering more robust targets for radiometric calibration efforts.

A multi-scale assessment of desertification severity in the Eastern Mediterranean region based on albedo-MSAVI feature space.

Resampling the same satellite image to conduct a multi-scale assessment of desertification can be accompanied by distortion of terrestrial objects and spectral information, which can lead to uncertainty in the generated information. To address this, this study assesses desertification severity in an area of arid and semi-arid climate in the Eastern Mediterranean (Jordan)that is characterised by cloudless scenesusingmulti-sensor data of the same scene at the same time. To this end, Sentinel-2 at 10 m and 60 m, Landsat-8 at 30 m and MODIS at 250 m and 500 m were collected to extract albedo and modified soil adjusted vegetation index (MSAVI), and subsequently to construct albedo-MSAVI feature space. Using the negative correlation between albedo and MSAVI, desertification degree index (DDI) was generated. The resulting multi-scale DDI maps bear a relative resemblance in terms of spatial distribution, patterns, and proportions. The DDI maps indicate that extremely serious and serious desertification are widespread, accounting for 50% of the study area, primarily in the eastern portions. However, finer DDI maps (10 m, 30 m and 60 m) are essential for detecting small-scale desertification characteristics due to their ability to capture local spatial variabilities, while coarser ones (250 m and 500 m) are better suited for capturing broad-scale desertification patterns driven by climatic factors, in which MODIS data exhibit a relatively higher positive correlation with seasonal average precipitation. Although finer DDI maps show higher accuracy compared to coarser ones, the accuracy of DDI maps of MODIS has shown an increase within a homogeneous landscape. Accordingly, synchronised multi-scale assessment of desertification severity is not only influenced by the spatial resolution but also by the landscape heterogeneity and the type of satellite sensor utilised. The multi-scale approach applied in this study can provide insights on scale-dependent desertificationthat help in devising overarching mitigation strategies.

Leveraging Social Media Data for Forest Fires Sentiment Classification: A Data-Driven Method

Background: The rise in forest fires over the last two years, which is due to rise in dry weather conditions and human activities, have greatly impacted an area of 1.6 million hectares, leading to significant ecological, economic, and health issues, hence the need to improve disaster response strategies. Previous research determined the lack of coverage regarding public response during forest fires with conventional methods such as satellite images and sensor data. However, social media platforms provide real-time information generated by users, along with location information of disaster events. Sentiment analysis helps in understanding the public reactions and responses to natural disasters, thereby increasing awareness about forest fires. Objective: The purpose of this research is to assess the efficiency of Long Short-Term Memory (LSTM) method in classifying sentiment for social networks in regard to forest fires. This research aims to examine the effect of TF-IDF, unigram, and the FastText features on the effectiveness of the classification of sentiment. Methods: The precision, recall, and F1 score of 2, 3, and 4 determined in the LSTM models with commonly available sentiment analysis tools, such as the Vader Sentiment Analysis and SentiWordNet was used to evaluate the performance of the model. Results: With an improvement of roughly 10%, the four layers of the LSTM model generated the best performance for the evaluation of sentiments about forest fires. The LSTM model with FastText achieved F1, recall and precision scores of 0.649, 0.641, and 0.659, which exceeds lexicon-based method including SentiWordNet and Vader. Conclusion: The experimental results showed that the LSTM model outperformed lexicon-based methods when used to analyse the tweets related to forest fire. Additional research is required to integrate rule-based models and LSTM models to develop a more robust model for dynamic data. Keywords: Forest Fire, Disaster, Long Short-Term Memory, LSTM, Vader, SentiWordnet

Shifts in Maritime Trade Routes as a Result of Red Sea Shipping Crisis Detected in TROPOMI NO2 Data

AbstractObservations from space‐borne spectrometers have been lately used to quantify shipping emissions of nitrogen oxides (ΝΟΧ). Here we present a method that enhances the shipping signal of NO2 TROPOspheric MOnitoring Instrument (TROPOMI) satellite sensor observations in order to assess the impact of the Red Sea ship attacks on NO2 levels in three important shipping routes along the Red Sea, the Cape of Good Hope, and the Gibraltar Strait. Major shipping carriers, sailing usually via the Red Sea, have responded to the attacks by transiting their fleet around the African continent. The shipping signal from TROPOMI declines by ∼55% in the Red Sea and ∼15% in the Gibraltar Strait while an increase of ∼40% is found off the South African coast between January–June 2024 and the same period in 2023. These changes correlate well with vessel statistics, demonstrating the ability to track abrupt changes in NO2 shipping levels with satellite measurements.

Development of VIIRS-OLCI chlorophyll-a product for the coastal estuaries

Coastal waters require monitoring of chlorophyll-a concentration (Chl-a) in a wide range of Chl-a from a few mg/m3 to hundreds of mg/m3, which is of interest to the fisheries industry, evaluation of climate change effects, ecological modeling and detection of Harmful Algal Blooms (HABs). Monitoring can be carried out from the Visible Infrared Imaging Radiometer Suite (VIIRS) and Ocean and Land Colour Instrument (OLCI) Ocean Color (OC) satellite sensors, which are currently on orbit and are expected to be the main operational OC sensors at least for the next decade. A Neural Network (NN) algorithm, which uses VIIRS M3-M5 reflectance bands and an I1 imaging band, was developed to estimate Chl-a in the Chesapeake Bay, for the whole range of Chl-a from clear waters in the Lower Bay to extreme bloom conditions in the Upper Bay and the Potomac River, where Chl-a can be used for bloom detection. The NN algorithm demonstrated a significant improvement in the Chl-a retrieval capabilities in comparison with other algorithms, which utilize only reflectance bands. OLCI NIR/red 709/665 nm bands red edge 2010 algorithm denoted as RE10 was also explored with several atmospheric corrections from EUMETSAT, NOAA and NASA. Good consistency between the two types of algorithms is shown for the bloom conditions and the whole range of waters in the Chesapeake Bay (with RE10 switch to OC4 for lower Chl-a) and these algorithms are recommended for the combined VIIRS-OLCI product for the estimation of Chl-a and bloom monitoring. The algorithms were expanded to the waters in Long Island Sound, demonstrating good performance.

Spatial Resolution Enhancement Framework Using Convolutional Attention-Based Token Mixer

Spatial resolution enhancement in remote sensing data aims to augment the level of detail and accuracy in images captured by satellite sensors. We proposed a novel spatial resolution enhancement framework using the convolutional attention-based token mixer method. This approach leveraged spatial context and semantic information to improve the spatial resolution of images. This method used the multi-head convolutional attention block and sub-pixel convolution to extract spatial and spectral information and fused them using the same technique. The multi-head convolutional attention block can effectively utilize the local information of spatial and spectral dimensions. The method was tested on two kinds of data types, which were the visual-thermal dataset and the visual-hyperspectral dataset. Our method was also compared with the state-of-the-art methods, including traditional methods and deep learning methods. The experiment results showed that the method was effective and outperformed state-of-the-art methods in overall, spatial, and spectral accuracies.

DSM Reconstruction from Uncalibrated Multi-View Satellite Stereo Images by RPC Estimation and Integration

In this paper, we propose a 3D Digital Surface Model (DSM) reconstruction method from uncalibrated Multi-view Satellite Stereo (MVSS) images, where Rational Polynomial Coefficient (RPC) sensor parameters are not available. While recent investigations have introduced several techniques to reconstruct high-precision and high-density DSMs from MVSS images, they inherently depend on the use of geo-corrected RPC sensor parameters. However, RPC parameters from satellite sensors are subject to being erroneous due to inaccurate sensor data. In addition, due to the increasing data availability from the internet, uncalibrated satellite images can be easily obtained without RPC parameters. This study proposes a novel method to reconstruct a 3D DSM from uncalibrated MVSS images by estimating and integrating RPC parameters. To do this, we first employ a structure from motion (SfM) and 3D homography-based geo-referencing method to reconstruct an initial DSM. Second, we sample 3D points from the initial DSM as references and reproject them to the 2D image space to determine 3D–2D correspondences. Using the correspondences, we directly calculate all RPC parameters. To overcome the memory shortage problem while running the large size of satellite images, we also propose an RPC integration method. Image space is partitioned to multiple tiles, and RPC estimation is performed independently in each tile. Then, all tiles’ RPCs are integrated into the final RPC to represent the geometry of the whole image space. Finally, the integrated RPC is used to run a true MVSS pipeline to obtain the 3D DSM. The experimental results show that the proposed method can achieve 1.455 m Mean Absolute Error (MAE) in the height map reconstruction from multi-view satellite benchmark datasets. We also show that the proposed method can be used to reconstruct a geo-referenced 3D DSM from uncalibrated and freely available Google Earth imagery.

Swincloud: a hybrid network for cloud detection in thermal infrared remote sensing images

ABSTRACT Cloud cover is a significant factor affecting the effectiveness of satellite-based Earth observations. Existing cloud detection algorithms primarily rely on imaging data from satellite sensors in the visible to near-infrared spectral range, making it challenging to achieve day-and-night cloud monitoring. Convolutional neural networks have shown outstanding performance in previous cloud detection algorithms due to their robust ability to extract local information. However, their inherent inductive bias limits their capacity to learn long-range semantic information. To address these challenges, we proposed SwinCloud, a U-shaped semantic segmentation network based on an enhanced Swin Transformer for cloud detection in the thermal infrared spectral range. Specifically, we augment the Swin Transformer's window attention module with a CNN-based parallel pathway to effectively model global-local information. We employ a feature fusion module before the final upsampling module in the decoder to better integrate low-level spatial information and high-level semantic information. On the Landsat-8 cloud detection dataset, our model outperforms state-of-the-art methods. When transferred to the SDGSAT-TIS cloud detection dataset, the mIOU of experiment results reaches 69.9%, demonstrating the strong transferability of SwinCloud across different sensors. We also applied SwinCloud to cloud detection in the visible bands of Landsat-8. The results demonstrated SwinCloud's generalization capability across different bands.

A New Generation of Satellite Sensors Based on Graphene and Carbon Nanotubes: A Review

A New Generation of Satellite Sensors Based on Graphene and Carbon Nanotubes: A Review

Accuracy Assessment of NOAA IMS 4 km Products on the Tibetan Plateau with Landsat-8 OLI Images

The NOAA IMS (Interactive Multisensor Snow and Ice Mapping System) is a blended snow and ice product based on active and passive satellite sensors, ground observation, and other auxiliary information, providing the daily cloud-free snow cover extent in the Northern Hemisphere (NH) and having great application potential in snow cover monitoring and research in the Tibetan Plateau (TP). However, accuracy assessment of products is crucial for various aspects of applications. In this study, Landsat-8 OLI images were used to evaluate and validate the accuracy of IMS products in snow cover monitoring on the TP. The results show that (1) average overall accuracy of IMS 4 km products is 76.0% and average mapping accuracy is 88.3%, indicating that IMS 4 km products are appropriate for large-scale snow cover monitoring on the TP. (2) IMS 4 km products tend to overestimate actual snow cover on the TP, with an average commission rate of 45.4% and omission rate of 11.7%, and generally present that the higher the proportion of snow-covered area, the lower the probability of omission rate and the higher the probability of commission rate. (3) Mapping accuracy of IMS 4 km snow cover on the TP generally is higher at the high altitudes, and commission and omission errors increase with the decrease of elevation. (4) Compared with less regional representativeness of ground observations, the spatial characteristics of snow cover based on high-resolution remote sensing data are much more detailed, and more reliable verification results can be obtained. (5) In addition to commission and omission error metrics, the overall accuracy and mapping accuracy based on the reference image instead of classified image can better reveal the general monitoring accuracy of IMS 4 km products on the TP area.

Cross-resolution land cover classification using outdated products and transformers

ABSTRACT Large-scale, high-resolution land cover classification is a prerequisite for constructing Earth system models and addressing ecological and resource issues. Advancements in satellite sensor technology have led to improvements in spatial resolution and wider coverage areas. Nevertheless, the lack of high-resolution labelled data is still a challenge, hindering the large-scale application of land cover classification methods. In this study, a Transformer-based weakly supervised method for cross-resolution land cover classification using outdated data is proposed. First, to capture long-range dependencies without overlooking the fine-grained details of objects, a U-Net-like Transformer based on a reverse difference mechanism (RDM) using dynamic sparse attention is designed. Second, an anti-noise loss calculation module based on optimal transport (OT) is proposed. The anti-noise loss calculation identifies confident areas and vague areas based on the OT matrix, which relieves the effect of noises on outdated land cover products. By introducing a weakly supervised loss with weights and using an unsupervised loss, the RDM-based U-Net-like Transformer was trained. Remote sensing images with 1 m resolutions and the corresponding ground truths of six states in the United States were used to validate the performance of the proposed method. The experiments used outdated land cover products with 30 m resolutions from 2013 as training labels and produced land cover maps with 1 m resolutions from 2017. The results showed the superiority of the proposed method over state-of-the-art methods. The code is available at https://github.com/yu-ni1989/ANLC-Former.

Synergistic Potential of Optical and Radar Remote Sensing for Snow Cover Monitoring

This research studies the characteristics of snow-covered area (SCA) from two vastly different sensors: optical (Moderate-Resolution Imaging Spectroradiometer, or MODIS, equipped on board the Terra satellite) and radar (Synthetic Aperture Radar (SAR) on-board Sentinel-1 satellites). The focus are the five mountain ranges of the Iberian Peninsula (Cantabrian System, Central System, Iberian Range, Pyrenees, and Sierra Nevada). The MODIS product was selected to identify SCA dynamics in these ranges using the Probability of Snow Cover Presence Index (PSCPI). In addition, we evaluate the potential advantage of the use of SAR remote sensing to complete optical SCA under cloudy conditions. For this purpose, we utilize the Copernicus High-Resolution Snow and Ice SAR Wet Snow (HRS&I SWS) product. The Pyrenees and the Sierra Nevada showed longer-lasting SCA duration and a higher PSCPI throughout the average year. Moreover, we demonstrate that the latitude gradient has a significant influence on the snowline elevation in the Iberian mountains (R2 ≥ 0.84). In the Iberian mountains, a general negative SCA trend is observed due to the recent climate change impacts, with a particularly pronounced decline in the winter months (December and January). Finally, in the Pyrenees, we found that wet snow detection has high potential for the spatial gap-filling of MODIS SCA in spring, contributing above 27% to the total SCA. Notably, the additional SCA provided in winter is also significant. Based on the results obtained in the Pyrenees, we can conclude that implementing techniques that combine SAR and optical satellite sensors for SCA detection may provide valuable additional SCA data for the other Iberian mountains, in which the radar product is not available.

Aerosol optical depth retrieval using scaled digital number (DN) values of multi-spectral satellite and a generating adversarial model based on deep learning application

ABSTRACT The current quantitative retrieval of Aerosol Optical Depth (AOD) typically uses Top-of-atmosphere (TOA) reflectance data obtained by the scale and offset terms of radiometric calibration. Errors can be introduced during the conversion of Digital Number (DN) values to TOA reflectance, restraining the accurate retrieval of AOD. Particularly when surface reflectance is high, conversion errors can significantly distort the retrieval of Aerosol Optical Depth (AOD), given the minimal impact of aerosols on the radiation detected by satellite sensors. To counteract this, we introduce an innovative retrieval algorithm that harnesses a Generative Adversarial Network (GANN), a deep learning (DL) model, to accurately derive AOD from the scaled digital number (DN) values of multi-spectral satellite imagery. The DL technology enables the use of scaled DN values for AOD retrieval, bypassing the conversion errors associated with TOA reflectance since it relies on a sample learning approach instead of radiative transfer calculations. To ensure the number and representativeness of training samples, a total of 37,103 samples from different underlying surfaces from 2014 to 2018 were obtained using a space-time matching strategy. Our K-cross validation demonstrates that employing DN values for AOD retrieval promises to yield higher accuracy compared to using TOA reflectance. Independent 2019 data were used to estimate GANN, MCD19A2 and MOD04_3K aerosol products. Our model showed superior performance compared to the other products in terms of a higher correlation (R = 0.8184) with AERONET AOD and obtaining more reliable retrievals (7601). The sensitivity experiment suggests that GANN exhibits limitations in areas characterized by high aerosol loads and heterogeneity. Our study can give a novel insight into AOD retrieval based on multi-spectral satellite and DL models.