Cloud Detection over Snow- or Ice-covered Surfaces Using Oxygen A-Band Observations

Cloud detection over snow- or ice-covered (S/IC) surfaces remains a critical challenge in satellite remote sensing. The cloud-like high surface albedo and ice-cloud-like brightness temperatures of these surfaces often lead to systematic misclassification in visible- and infrared-based algorithms, including the threshold-based cloud detection applied to Sentinel-5P atmospheric composition retrievals. Misclassified clear-sky scenes can introduce biases in the retrieved total columns of ozone, sulfur dioxide, and nitrogen dioxide, while misclassified cloudy scenes reduce the spatial coverage of satellite products.

To address this challenge, we develop a global cloud detection algorithm based on absorption images derived from Sentinel-5P oxygen A-band observations. The algorithm exploits the reduction of oxygen absorption in cloudy pixels, as cloud layers reflect solar radiation before it reaches the underlying surface, thereby shortening the radiative transfer path in the atmosphere and reducing absorption along the path. In addition, spatial texture information extracted from oxygen absorption images is incorporated to enhance sensitivity to optically thin and broken clouds, enabling more robust discrimination between clouds and bright underlying surfaces. This physical mechanism makes the algorithm insensitive to surface type, rendering it particularly suitable for global cloud detection, including over S/IC surfaces.

Validation against CALIPSO demonstrates a marked improvement in cloud detection performance across diverse surface and cloud conditions. The proposed algorithm achieves an overall accuracy of 91%, compared with 85% for the Suomi-NPP product and 48% for the operational Sentinel-5P product. Improvements are particularly pronounced over S/IC surfaces, where detection accuracy increases by 15% relative to Suomi-NPP. Additionally, detection accuracy for optically thin clouds improves by 20% globally, with the largest gains (up to 52%) observed over S/IC surfaces. These results demonstrate the value of oxygen absorption and spatial texture features for cloud detection, especially over S/IC surfaces, and support improved quality and consistency of satellite-based atmospheric observations over polar and other bright-surface regions.