29 March 2021 in Atmospheric Chemistry and Physics
To better understand the aerosol properties over the Arctic, Antarctic and Tibetan Plateau (TP), the aerosol optical properties were investigated using 13 years of CALIPSO (Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations) L3 data, and the back trajectories for air masses were also simulated using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT) model. The results show that the aerosol optical depth (AOD) has obvious spatial- and seasonal-variation characteristics, and the aerosol loading over Eurasia, Ross Sea and South Asia is relatively large. The annual-average AODs over the Arctic, Antarctic and TP are 0.046, 0.024 and 0.098, respectively. Seasonally, the AOD values are larger from late autumn to early spring in the Arctic, in winter and spring in the Antarctic, and in spring and summer over the TP. There are no significant temporal trends of AOD anomalies in the three study regions. Clean marine and dust-related aerosols are the dominant types over ocean and land, respectively, in both the Arctic and Antarctic, while dust-related aerosol types have greater occurrence frequency (OF) over the TP. The OF of dust-related and elevated smoke is large for a broad range of heights, indicating that they are likely transported aerosols, while other types of aerosols mainly occurred at heights below 2 km in the Antarctic and Arctic. The maximum OF of dust-related aerosols mainly occurs at 6 km altitude over the TP. The analysis of back trajectories of the air masses shows large differences among different regions and seasons. The Arctic region is more vulnerable to mid-latitude pollutants than the Antarctic region, especially in winter and spring, while the air masses in the TP are mainly from the Iranian Plateau, Tarim Basin and South Asia.
20 September 2023 in IEEE Transactions on Geoscience and Remote Sensing
In this study, a thermal infrared (TIR) based convolutional neural network (TIR-CNN), originally designed for retrieving cloud optical properties from TIR radiometry, is further developed to obtain global cloud base height (CBH) and cloud thermodynamic phase during both daytime and nighttime. It employs TIR radiances, cloud optical properties retrieved by TIR-CNN, altitude, landcover, and lifting condensation level as inputs to estimate global CBH and cloud phase for both single- and multi-layer clouds. This new model is trained, validated, and tested using radar-lidar products from CloudSat/CALIPSO. It provides global CBH with root-mean-square errors of 1.19 km and 1.91 km for single- and multi-layer clouds, respectively. A cloud layer classifier is trained to provide information on the quality of retrieved CBH, with total accuracies of 82% and 85% for single- and multi-layer clouds, respectively. In addition, the new model has remarkable accuracy in identifying the cloud phase within each pixel’s vertical column, particularly in differentiating mixed-phase clouds with an ice cloud top.
20 September 2018 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
In terms of traditional threshold methods, uniform thresholds are used for cloud detection based on remote sensing images; however, due to complex surface structures and cloud conditions, such an approach is typically difficult to effectively implement for high-precision cloud detection. To solve this problem, a new cloud detection algorithm is proposed based on global land cover data. Specifically, a high spatial-resolution at 30-m Global Land Cover Data set with global coverage was employed as background data for image inversions, which further supported cloud detection in remote sensing images. Notably, threshold settings can be varied for different land cover types. Such an algorithm can effectively improve the accuracy of cloud pixel identification for thin and broken clouds, even over bright areas. Moreover, Landsat 5 data are used to perform cloud detection experiments based on this algorithm. The thresholds are considering land cover variations. The thresholds of land cover types spatiotemporally vary, such as vegetation, differed by latitude and over time. In addition, six common land cover types are selected for cloud detection experiments. Then, validations analyses are conducted through visual interpretation and the results indicated that the algorithm is capable of achieving a high cloud detection accuracy. Specifically, the overall RMSE of cloud cover is 4.44%, and the accuracies of cloud and clear-sky pixel identifications is 86.5% and 98.7%, respectively.
30 November 2024 in Journal of Geophysical Research: Machine Learning and Computation
Marine low clouds play a crucial role in cooling the climate, but accurately predicting them remains challenging due to their highly non‐linear response to various factors. Previous studies usually overlook the effects of cloud droplet number concentration (Nd) and the non‐local information of the target grids. To address these challenges, we introduce a convolutional neural network model (CNNMet‐Nd) that uses both local and non‐local information and includes Nd as a cloud‐controlling factor to enhance the predictive ability of daily cloud cover, albedo, and cloud radiative effects (CRE) for global marine low clouds. CNNMet‐Nd demonstrates superior performance, explaining over 70% of the variance in these three cloud variables for scenes of 1° × 1°, a notable improvement over past efforts. CNNMet‐Nd also accurately replicates geographical patterns of trends in marine low clouds from 2003 to 2022. In contrast, a similar model without Nd (CNNMet) struggles to predict long‐term trends in cloud properties effectively. Permutation importance analysis further highlights the critical role of Nd in CNNMet‐N's predictive success. Further comparisons with an artificial neural network (ANNMet‐Nd) model, which uses the same inputs but without considering spatial dependence, show CNNMet‐Nd's superior performance with R2 values for cloud cover, albedo, and CRE being 0.16, 0.12, and 0.18 higher, respectively. This highlights the importance of incorporating non‐local information, at least on a daily scale, into low cloud predictions to enhance climate model parameterizations.
4 July 2018 in IEEE Geoscience and Remote Sensing Letters
Prior knowledge of the background land surface reflectance (LSR) constitutes one of the most important factors affecting the precision of cloud shadow detection. To resolve this problem, a surface reflectance-based cloud shadow detection (SRCSD) algorithm is proposed for multitemporal Landsat images. Monthly surface reflectance data sets constructed from MODIS surface reflectance products (MOD09A1) were used to provide the background LSR for cloud shadow detection. Based on the background LSR, the possible variation in the top of atmosphere (TOA) reflectance for each clear pixel can be estimated using the radiative transfer equation under different atmospheric conditions. If a pixel has a smaller TOA reflectance than the minimum value of the possible range under clear conditions, it is identified as being shadow covered. One hundred and twenty-five Landsat 8 Operational Land Imager scenes covered by various surface types were selected to evaluate the feasibility of the algorithm. A validation using manual cloud shadow masks showed that the average producer’s accuracy and user’s accuracy were approximately 0.805 and 0.893, respectively. A comparison of the results of the SRCSD algorithm with those of an object-based cloud shadow detection algorithm (Fmask) recently developed for Landsat images revealed that SRCSD generally detects cloud shadows better than Fmask. The most significant improvement of the SRCSD algorithm is the better detection capability for thin and broken cloud shadows, and this algorithm can be extended to multiple types of satellite data after proper modification.