This study presents a deep-learning (D2S) model to generate historical vertically polarized (TBv) and horizontally polarized (TBh) brightness temperatures (TBs) at the Soil Moisture Active-Passive (SMAP) 1.4 GHz channels using TBv and TBs polarization difference from the Defense Meteorological Satellite Program (DMSP) satellite series' 19.35 GHz channels. The proposed D2S model employed the ResNet9-based generator and PatchGAN-based discriminator, with a compositve loss function synergizing mean squared error and L1 losses, similar to the Pix2Pix architecture. Training, validation, and testing were conducted using paired datasets of polarized TBs from SMAP (1.41 GHz) and DMSP (19.35 GHz) collected between July 2015 and June 2022, followed by application to historical data spanning from September 1987 to June 2015. The D2S model demonstrated robust performance in generating both SMAP TBv and TBh, achieving test set average correlation coefficients of 0.994 and 0.992, biases of 0.150 K and -0.259 K, root-mean-square errors of 6.463 K and 9.011 K, and mean absolute errors of 3.883 K and 5.518 K, respectively. Furthermore, this study showed reasonable agreement between the D2S-generated SMAP TBs and SMOS TBs and provided preliminary results of thin sea ice thickness retrievals, which were further used for evaluating the limitations of the D2S approach in specific sea ice regions. Consequently, the D2S model can provide empirically consistent SMAP-like polarized TB estimates over the Arctic, including periods prior to the L-band operational era. The D2S framework offers significant potential to enhance the reconstruction of historical Arctic sea ice conditions and to support climate change studies by providing surrogate L-band–equivalent information extending back to 1987.

Rainfall rate (RR) retrieval methods utilizing visible (VIS), near-infrared (NIR), and infrared (IR) bands from geostationary (GEO) meteorological satellites often achieve higher accuracy compared to methods relying solely on IR bands. However, these methods face limitations for nighttime applications. This study aims to introduce an artificial intelligence (AI)-based approach for nighttime RR retrieval using actual IR observations and virtual nighttime VIS (0.47, 0.51, and 0.86 μm) and NIR (1.37 and 1.61 μm) bands data generated through an adversarial data-to-data (D2D) translation method. The study focuses on the Korean Peninsula and employs paired observational datasets from the Advanced Meteorological Imager (AMI) on Geo-KOMPSAT-2A (GK2A) and hybrid surface rainfall data from ground weather radars as input and target domains for the D2D model. The study generates virtual D2D-Generated (DG) VIS and NIR band data and evaluates DG-RR using only IR bands during nighttime. Results indicate that DG-RR using virtual AMI VIS, NIR, and real AMI IR bands demonstrates superior quantitative and qualitative statistical scores compared to DG-RR based solely on the AMI IR band data. This improvement is observed over traditional algorithm-based GK2A AMI RR and precipitation estimation from remotely sensed information using artificial neural networks. Thus, this study underscores the potential of utilizing virtual AI-based VIS and NIR band data, resembling actual observations, in RR estimation and various satellite remote sensing applications.

This study presents a deep learning (DL)-based data-to-data (D2D) translation framework for generating synthetic Advanced Microwave Scanning Radiometer-2 (AMSR2)-like dual-polarization (Pol) measurements from the Global Precipitation Measurement Microwave Imager (GMI) 23.8 GHz vertical (V)-Pol data. The proposed D2D model was constructed through two D2D translations: AMSR2 V-Pol to AMSR2 horizontal (H)-Pol (main model) and GMI V-Pol to AMSR2 V-Pol (submodel). The D2D method incorporates a normalization preprocess and a denormalization postprocess and utilizes an adversarial learning framework for interdomain conversion of the physical values in brightness temperature ($T_{B}$) data. The datasets from AMSR2 at 23.8 GHz dual-Pol measurements, covering the period from January 2013 to December 2022 and from GMI at 23.8 GHz V-Pol measurements (23.8V GHz), covering the period from January 2015 to December 2022, were employed as the source and target datasets for training and evaluating the D2D model. The D2D-generated$T_{B}$data were validated against the AMSR2 observation data using statistical metrics, including the correlation coefficient (CC), bias, mean absolute error (MAE), and root mean square error (RMSE). The D2D-generated AMSR2 23.8 GHz H-Pol measurements (23.8H GHz) showed average values of CC =0.989, bias =0.8 K, MAE =4.0 K, and RMSE =5.8 K when validated against target data. Furthermore, this study demonstrated the critical role of H-Pol measurements in cloud liquid water (CLW) estimation experiments. Consequently, the hypothetical AMSR2 23.8 GHz dual-Pol data can provide additional valuable atmospheric information, thereby enhancing weather forecasting accuracy and climate change studies.

Accurate cloud mask (CM) information is essential for distinguishing between cloud-free and cloudy pixels in various satellite remote sensing applications. This study presents a daytime cloud detection method for the Advanced Meteorological Imager (AMI) sensor on board the Geo-Kompsat 2A satellite. The proposed cloud detection algorithm utilizes the AMI’s four bands (0.51 μm, 0.86 μm, 1.38 μm, and 1.61 μm) in combination with Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) observations. Thick clouds are detected using a conversion relationship between the green band (0.51 μm) and the normalized difference water index at the 0.51 μm and 0.86 μm bands, while thin clouds are detected using the 1.38 μm and 1.61 μm bands with empirically-determined threshold values between the collocated AMI and CALIPSO observations. A few overestimated cloud pixels are corrected using the normalized difference snow index, which consists of reflectance values at 0.51 μm and 1.61 μm. Case studies were performed in the East Asia region, including Korea, Japan, and the southeastern part of China, for the four seasons from 2020 to 2021. The proposed cloud detection method was validated using CALIPSO Vertical Feature Mask data. Results showed excellent statistical scores: probability of detection (POD) = 0.92, false alarm ratio (FAR) = 0.11, and proportion correct (PC) = 0.87 for 2020 cases, and POD = 0.92, FAR = 0.11, and PC = 0.86 for 2021 cases. Moreover, the proposed method demonstrated the significant benefits of distinguishing clouds from sea ice and snow over land in winter.

Sea surface salinity (SSS) is a key parameter in physical oceanography, global hydrological, biochemical cycles, and climate change studies. L-band radiometers onboard satellites, such as Soil Moisture Active Passive (SMAP), Soil Moisture and Ocean Salinity (SMOS), have played a crucial role in monitoring and analyzing the global SSS in the past decades. This study presents a method to retrieve and predict the SMAP SSS through data-to-data translation (D2D) based on conditional generative adversarial networks. The model was constructed from the polarized brightness temperatures, differences in the polarized brightness temperatures from the L band of the SMAP satellite, and sea surface temperature and sea surface wind speed from the European Center for Medium-Range Weather Forecasts data from April 2015 to July 2020, and applied to produce SMAP SSS. A comparison between the SMAP salinity and the D2D-generated SMAP salinity showed excellent agreement, evaluated through bias = 0.016, root mean square error (RMSE) = 0.173 in practical salinity unit (psu) units, and correlation coefficient (CC) = 0.985. Furthermore, a comparison between the D2D-generated and buoy-observed salinities showed good agreement (bias = -0.031 psu and RMSE = 0.196 psu, CC = 0.971). Additionally, the results of the one-month prediction model were also in good agreement with the SMAP SSS (bias = 0.028 psu and RMSE = 0.218 psu, CC = 0.977). Consequently, the D2D-based model can be effectively used to generate SMAP SSS information and can be applied to various microwave satellites.

This study proposes a deep-learning-based data-to-data (D2D) translation framework to simulate a radar-like retrieval of rainfall rates using the advantages of spatial coverage and temporal resolution of geostationary (GEO) satellite observation. The D2D method comprises normalization and denormalization in pre- and post-processing and an adversarial learning structure for an inter-domain conversion between physical values of data such as albedo and brightness temperature (BT) unlike the image-to-image translation using digital number values in image data. The GEO-KOMPSAT-2A (GK2A) and radar hybrid surface rainfall (HSR) datasets over the Korean Peninsula from September 2019 to September 2021 were used as the source and target domains for training and testing the D2D model. The constructed D2D model for ground radar-like rainfall generation was validated using the ground radar-observed rainfall data and compared to the GK2A rainfall rate (RR), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks-Cloud Classification System (PERSIANN-CCS), and Integrated Multi-satellite Retrievals for Global Precipitation Measurement (IMERG) rainfall products. The D2D model exhibited excellent performance for various rain types in the study area compared to the GK2A RR, PERSIANN-CCS, and IMERG data. Consequently, the D2D model can provide valuable and accurate radar-like rainfall intensity and distribution data with a high temporal resolution and complementary rainfall information over lands and oceans without radar observation.

Weather radars play an important role inin siturainfall monitoring owing to their ability to measure instantaneous rain rates and rainfall distributions. Currently, the Korea Meteorological Administration (KMA) provides instantaneous radar observation data and predictions based on the McGill algorithm for precipitation nowcasting by Lagrangian extrapolation (MAPLE) for up to 6 h, for short-term forecasting. This study presents a conditional generative adversarial network (CGAN)-based radar rainfall prediction method for very short-range weather forecasts from 10 min to 4 h. The CGAN-predicted model was trained and tested using KMA’s constant altitude plan position indicator (CAPPI) observation data. The qualitative comparison between the radar observation and the CGAN-predicted rain rates displayed high statistical scores, such as the probability of detection (POD) = 0.8442, false alarm ratio (FAR) = 0.2913, and critical success index (CSI) = 0.6268, in the case of a 1-h prediction for rainfall on September 5, 2019, 15:20 KST. This study demonstrates the capability of the CGAN model for short-term rainfall forecasting. Consequently, the CGAN-generated radar-based rainfall prediction could complement the KMA MAPLE system and be useful in various forecasting applications.

Green bands in satellite remote sensing play an important role in monitoring water and vegetation information. Due to the lack of observed green band, the Geostationary Operational Environmental Satellite (GOES-16) Advanced Baseline Imager (ABI) sensor uses a synthetic one. This study presents an ABI green band generation method using the Pix2Pix based on conditional generative adversarial networks (CGANs) and convolutional neural network techniques with data observed in the visible range of the GOES-16/ABI sensor. Our model was constructed from the radiance data sets in the red, blue, and green bands of the Advanced Himawari Imager (AHI) onboard Himawari-8/9 satellites from August 27, 2018 to May 1, 2019, and applied to generate a GOES-16 ABI green band using the ABI blue band radiance data. A comparison between the AHI and the Pix2Pix-generated AHI green bands displayed high accuracy, evaluated through bias = 0.120, root mean square error (RMSE) = 0.983 in digital number (DN) units, and correlation coefficient (CC) = 0.999. Furthermore, comparison between the Pix2Pix-generated and synthetic ABI green bands resulted in a good agreement (bias = 1.029 and RMSE = 2.892 in DN units, CC = 0.993). The statistical comparison between the green band, and red or blue band resulted in the exceptional performance of the Pix2Pix-generated ABI green band compared to the synthetic ABI green band. Consequently, our Pix2Pix-based model can be effectively used to generate nonexistent green band of ABI sensor and be applied in a variety of scientific applications requiring green band.

Certain characteristics of a material such as the surface reflectivities can be determined even without knowledge of its internal properties. In this letter, a direct relationship (Azzam relationship) and an analytical approximation (Azzam-Sohn-Hong (ASH) approximation) between the vertically and horizontally polarized reflectivities of specular surfaces are derived and validated using the refractive indices of water and metal in a variety of spectral regions. For the purpose of practical remote applications, land, sea water, sea ice, and oil surfaces are considered and compared using the Hong and ASH approximations. Consequently, ASH approximation has an advantage in a variety of spectral bands for materials with a small imaginary part of refractive index, while the Hong approximation does well in the microwave spectral region, or when the imaginary part of the reflective index is not neglected. Thus, a combination of the Hong and ASH approximations is suggested to improve upon previous studies that used only the Hong approximation in a variety of applications.

Currently, various radiative transfer (RT) schemes use their own number of layers in a plane-parallel atmospheric model to simulate the brightness temperature (TB) without considering the effect. The purpose of this letter is to understand numerically and analytically the TB uncertainty due to the layer number of the plane-parallel atmospheric model in RT schemes. Methodically, a plane-parallel atmospheric model in light rainfall and two RT schemes are used. The simulated TBs show a dependence on the number of layers in a plane-parallel atmosphere, which is due to the optical thickness. This letter suggests that the vertical resolution should be considered as a parameter in the forward model of RT schemes.