A Deep Learning Trained Clear-Sky Mask Algorithm for VIIRS Radiometric Bias Assessment

Abstract

Clear-sky mask (CSM) is a crucial influence on the calculating accuracy of the sensor radiometric biases for spectral bands of visible, infrared, and microwave regions. In this study, a fully connected deep neural network (FCDN) was proposed to generate CSM for the Visible Infrared Imaging Radiometer Suite (VIIRS) onboard Suomi National Polar-Orbiting Partnership (S-NPP) and NOAA-20 satellites. The model, well-trained by S-NPP data, was used to generate both S-NPP and NOAA-20 CSMs for the independent data, and the results were validated against the biases between the sensor observations and Community Radiative Transfer Model (CRTM) calculations (O-M). The preliminary result shows that the FCDN-CSM model works well for identifying clear-sky pixels. Both O-M mean biases and standard deviations were comparable with the Advance Clear-Sky Processor over Ocean (ACSPO) and were significantly better than a prototype cloud mask (PCM) and the case without a clear-sky check. In addition, by replacing CRTM brightness temperatures (BTs) with the atmosphere air temperature and water vapor contents as input features, the FCDN-CSM exhibits its potential to generate fast and accurate VIIRS CSM onboard follow-up Joint Polar Satellite System (JPSS) satellites for sensor calibration and validation before the physics-based CSM is available.

Applying Deep Learning to Clear-Sky Radiance Simulation for VIIRS with Community Radiative Transfer Model—Part 2: Model Architecture and Assessment

“…An FCDN is a multilayered artificial neural architecture, which is widely used among deep-learning models to solve problems of function fitting, classification, clustering, and pattern recognition. Liang et al [27] summarized the details of the FCDN, which was successfully applied to the classification problem of the VIIRS clear-sky mask for efficient and accurate O-M validation in global.…”

“…Therefore, the design of the FCDN_CRTM architecture was more complex to ensure rapid convergence and to attain a global optimum for the cost function (also called the loss function) during the model training. As discussed in [27], there is no mathematical or physical rule to determine the best hyperparameters, other than early ANN references and fine-tuning by repeated experiments. By effort in extensive experiments and model fine-tuning, we finally designed three hidden layers with 512, 384, and 64 neurons in the layers, respectively.…”

“…A new algorithm of the VIIRS clear-sky mask using the FCDN (FCDN_CSM) [27] was developed to replace the traditional physical-based model. The aim is to identify clear-sky domain efficiently for the real-time monitoring of VIIRS O-M biases in the ICVS system.…”

“…We present only the trends for the mean and STD for M15, but the performances of other bands were similar. As discussed in [27], the cost function, mean, and STD oscillated up and down during the training; this was due to using small batch sizes instead of a single sample in each iteration. Table 3 compares the F-C mean biases and STDs between the training, testing, and prediction data.…”

“…To explore the efficiency and accuracy of ANN application in the CRTM and in the real-time monitoring of sensor radiometric biases in global, we designed and developed a fully connected deep neural network (FCDN) algorithm and applied it to CRTM simulation for the Suomi-National Polar-orbiting Partnership (SNPP) VIIRS in five TEB/M bands. Together with the earlier-developed FCDN clear-sky mask (FCDN_CSM) [27,28], the objectives in this study are (1) to predict global clear-sky BTs using a well-trained FCDN_CRTM for high spatial resolution VIIRS in near real time and (2) to validate the FCDN_CRTM prediction accuracy, efficiency, and long-term stability. Section 2 discusses the methodology of this research.…”