2026-10-14T15:58:04Z https://www.scidb.cn/oai

10.57760/sciencedb.space.03571 2026-05-13T15:21:17Z

2026-05-13 A Dataset for the Fast Atmospheric Correction Model of L-band Spaceborne Microwave Radiometer Based on Neural Network doi:10.57760/sciencedb.space.03571 en This dataset is a supporting data product generated for the construction, training, and validation of a fast atmospheric correction model for L-band spaceborne microwave radiometers. Atmospheric correction is a critical step in sea surface salinity remote sensing using L-band microwave radiometers. Traditional statistical regression and parametric methods have limitations in terms of expressiveness and operational convenience. To address these issues, this study introduces a neural network approach to L-band microwave radiometer atmospheric correction for the first time, and proposes and constructs an A-B coefficient atmospheric correction model based on a BP neural network.This dataset was generated using the MPM93 atmospheric radiative transfer physical model and the ERA5 hourly reanalysis data from the European Centre for Medium-Range Weather Forecasts (ECMWF), covering clear-sky ocean areas within the global latitude range of 60°S to 60°N. The specific data generation and processing procedures are as follows: First, the traditional top-of-atmosphere brightness temperature model was mathematically transformed to derive a linear atmospheric correction equation with respect to the sea surface brightness temperature, where the slope and intercept of the equation are defined as the atmospheric correction coefficients A and B. Second, based on ERA5 surface data and 37-layer standard pressure level profile data, vertical profiles of atmospheric temperature, humidity, and pressure were obtained on a 60-layer non-uniform vertical grid through interpolation and physical formula calculations. Key input parameters, including surface temperature, surface pressure, surface water vapor density, and total column water vapor, were then derived. Third, the MPM93 model was used to perform radiative transfer integrations to calculate the upwelling brightness temperature, downwelling brightness temperature, and total atmospheric loss factor under different incidence angles, from which the corresponding true values of the A and B coefficients were derived. Finally, a sensitivity analysis of the input parameters and correction coefficients was conducted, with surface water vapor density selected as the primary humidity input variable. A 5-fold cross-validation method was employed to optimize the number of neurons in the hidden layer of the neural network, resulting in a standardized dataset suitable for model training, validation, and testing.The dataset covers a range of atmospheric and environmental parameters, including incidence angle (0°–70°), surface temperature (270–310 K), surface pressure (980–1040 hPa), surface water vapor density (0–30 g/m³), total column water vapor (0–80 kg/m²), and extraterrestrial radiation brightness temperature (0–20 K). The total number of sample records in this dataset is 9,924,380. Each record contains a complete set of atmospheric parameters and the corresponding true values of the A and B coefficients calculated using the MPM93 model. All data are numerical, with no missing values. Data errors mainly originate from estimation errors in the atmospheric parameters of the ERA5 reanalysis data and approximation errors of the MPM93 model under extreme atmospheric conditions; however, the overall error level is low, meeting the requirements for high-precision atmospheric correction modeling.The dataset is stored as a tabular file (e.g., in CSV or MAT format) containing the following fields: incidence angle, surface temperature, surface water vapor density, surface pressure, extraterrestrial radiation brightness temperature, true A coefficient, and true B coefficient. The data file structure is clear, with column labels corresponding to the abbreviations of each physical quantity. Row labels represent sample indices. All physical quantities use standard international units (angle in degrees, temperature in K, pressure in hPa, water vapor density in g/m³, brightness temperature in K). This dataset can be directly used for deep learning or statistical regression modeling of atmospheric correction models for L-band microwave radiometers, as well as for accuracy comparison and validation analyses of existing atmospheric correction models. L-band; microwave radiometer; neural network zhu bo wen PUBLIC https://creativecommons.org/licenses/by-nc/4.0/ dataset Science Data Bank