Accurate retrieval of the spatiotemporal distribution of atmospheric aerosols is essential for studying aerosol-radiation-cloud interactions, air-quality forecasting, and climate‑change assessment. Although data-driven methods have significantly advanced aerosol retrieval, existing models often neglect the influence of aerosol type on retrieval accuracy. To address this gap, this study presents an improved data- driven aerosol retrieval framework that explicitly incorporates aerosol type information into model training. Aerosol classification is performed using the K-means unsupervised clustering algorithm to optimize training samples, thereby enhancing model adaptability and retrieval accuracy. The refined samples are then used to train an Extremely Randomized Trees (ERT) model, achieving an optimal balance between accuracy and computational efficiency. Validation results demonstrate strong performance, with a correlation coefficient of 0.93, a root mean square error (RMSE) of 0.072, and over 89 % of results falling within the expected error range [(EE: ± (0.05 + 20 % × in-situ observations)], better than that of the traditional model. The findings demonstrate that integrating aerosol- type information into data- driven retrievals substantially improves accuracy and applicability for aerosol remote sensing. Future research should focus on refining aerosol classification techniques and integrating multi-source remote sensing data to enhance model robustness and global applicability further.

Although multispectral satellite remote sensing (RS) is able to derive relatedly accurate aerosols on a large geographical scale, which is crucial for the study of aerosol-related climate and environment changes, some issues remain in current retrieval algorithms. For example, both radiance transfer and machine learning algorithms fail to consider the aerosol types because it is challenging to use only multispectral information to quantify aerosol sources accurately. This may cause a large uncertainty in RS aerosol retrieval, especially in areas with complex and varying aerosol components. Moreover, there are multiple parameters that can reflect aerosol’s physical and chemical properties, but most developed algorithms only accurately obtain one once, such as aerosol optical depth (AOD), which may lead to a large inconsistency when applying them from different algorithms. To address these issues, we propose a multiparameter aerosol simultaneous retrieval algorithm by combining multispectral satellite and atmospheric simulation using a space-time transformer (STTF) model. Sample-based ten-fold validation suggests that our STTF model can effectively retrieve multiple aerosol parameters in terms of 550-nm AOD with R (root mean square error, RMSE) of 0.89 (0.10), Ångström exponent (AE) with R (RMSE) of 0.89 (0.26), and single scattering albedo (SSA) with R (RMSE) of 0.89 (0.10). The time- and spatial-based validation further underscores the model’s ability to predict different aerosol parameters over areas or periods without ground-based measurements. Moreover, the model also shows better AOD retrievals than the operational aerosol products (i.e., MCD19A2 and MOD04_3K) over the word and has significant improvement in areas with high aerosol loadings.

This paper proposes a generic algorithm to detect clouds in nighttime satellite images by using daytime image samples. The method aims to overcome the challenges of poor quality and inadequate number of nighttime image samples when detecting clouds based on machine learning (ML) during the night. Importantly, to precisely detect thin broken clouds and cloud edges during nighttime, we first use contextual features as input of the ML model to acquire more comprehensive information and enhance its ability to recognize these types of clouds. Nighttime images observed by the Gaofen-5 VIMI were utilized as an example to validate the performance of our algorithm. The result demonstrates that the addition of contextual features causes ML model to exhibit better performance and show a more accurate and refined detection of thin broken clouds and cloud edges, along with an overall accuracy of 91.38%, and a 3.64% improvement compared to the model without these features.

Nighttime cloud detection of satellite imagery encounters challenges due to the limited availability of bands during night. Hence, to tack the issue of acquiring high-quality samples for nighttime cloud detection, based on the analysis of the thermal infrared band (TIN) radiation transmission process and the fact that daytime and nighttime have the same source of radiance in TIN, we propose a generic nighttime cloud detection method, which involves using daytime long wave infrared (8-14μm) data to construct samples that can be used to train data-driven models specifically designed for nighttime cloud detection. To access the applicability of this method, data from MODIS (Moderate Resolution Imaging Spectroradiometer) satellite sensor was used to construct samples using the method and were applied for data-driven models. The results were validated by the Lidar cloud product and show that our nighttime cloud detection algorithm has higher accuracy compared to the official product (MYD35).

Deep learning excels in advanced inference tasks using electronic neural networks (ENN), but faces energy consumption and limited computation speed challenges. To mitigate this, optical neural networks (ONNs) were developed, utilizing light for computations. However, their high manufacturing costs limited accessibility. In this work, we first introduce the binary optical neural network (BONN) - a streamlined ONN variant with binarized weights, which significantly reduces fabrication complexities and costs. Specifically, we address (i) the development of a binarization weight function aligned with backward-error propagation, and (ii) a simulation-based training for extra-large neural networks housing millions of neurons. We prototype six BONNs, each comprising four 0.8 × 0.8mm2 layers with one million 800 nm diameter neurons. Costs are cut to 0.13 USD per layer, marking a substantial decrease of 769× from previous ONNs. Experimental results reveal BONNs consume 2, 405× less power than leading ENNs while maintaining an average recognition accuracy of 74% across six datasets.

Ambient fine particulate matter (PM2.5) has significant adverse effects on human health, thereby urgent hunger for accurate monitoring of ground-level PM2.5, especially its space distribution. Since satellites can observe the Earth on a large spatial scale, remote sensing technology can be applied to estimate PM2.5concentrations at the national level. Based on it and machine learning (ML) methods, numerous studies mapped high-accuracy, wholesale and continuous PM2.5. However, different models and data in these studies made their results incomparable, and more samples were needed to be provided. Here, a large-column and long-term sample dataset (ML-CNPM2.5) applied for ML-based models was constructed with 5,076,608 data records and 24 features from 2014 to 2023 in China. Multiple approaches were used to guarantee the quantity and quality of the sample dataset. Due to its comprehensiveness and objectivity, the ML-CNPM2.5can be used to train and validate different models, thereby further improving the accuracy of PM2.5estimating. Using the ML-CNPM2.5, eight basic ML-based models were also constructed as the baseline for judging other derivative models. These models can estimate daily full-coverage PM2.5and most performed well, with 10-fold cross-validation RMSE of 16.94-11.21μg/m3and R2of 0.71-0.89, which is consistent with previous studies and can effectively capture spatial trends of PM2.5in a period suffered from high pollution. Overall, our ML-CNPM2.5can be applied to effectively construct, validate, and compare various ML-based models for PM2.5estimation, helping to develop new algorithms with higher accuracy and robustness.

The signal or clutter separation in airborne MIMO SAR by the information of the direction-of-departure (DOD) or the direction-of-arrival (DOA) is able to be of benefit to the suppression of the “unwanted” signals. As the signal separation performance after using the blocking matrix algorithm is greatly affected by the estimate deviation of the direction. Moreover, the direction of the “unwanted” signal is unknown exactly in the practical application. To alleviate this situation, in this paper, a Modified Kernel Canonical Correlation Analysis (MKCCA) approach based on the blocking matrix approach is proposed. Firstly, the initial separation result is obtained by applying the blocking matrix approach. Then, we use KCCA as a further separation step. The azimuth searching result is adopted as the criteria of the separation performance. Simulation results demonstrate the excellent performance of the proposed approach, which could separate the signals or the clutter “cleaner”.