In this study, a new deep learning method was developed to estimate the spatiotemporal properties of the hourly aerosol optical depth (AOD) because existing physical models are limited in their abilities to separate the reflectance between aerosols and the underlying surface over land, accurately and effectively. By incorporating geostationary ocean color imagery (GOCI), multispectral bands were applied to train data-driven models to estimate the high-spatiotemporal-resolution AOD over Northeast Asia. Physical model and traditional machine learning (ML) models (the random forest (RF) and support vector regression (SVR) models) were compared with the deep neural network (DNN) model to evaluate its accuracy, implementing hold-out validation and $k$ -fold cross-validation approaches. In the statistical results of the hold-out validation, the DNN model showed the higher accuracy (root mean square error (RMSE) = 0.112, mean bias error (MBE) = 0.007, and correlation coefficient $(R) = 0.863$ ) relative to the traditional SVR (RMSE = 0.123, MBE = −0.010, and $R = 0.833$ ) and RF (RMSE = 0.125, MBE = 0.004, and $R = 0.825$ ) models. The DNN model also exhibited the best performance for most statistical metrics among the traditional SVR, RF, and selected physical models (except for the correlation coefficients and index of agreement) in the spatial and temporal cross-validation analyses. Although the DNN model was trained using the match-up dataset between the top of atmosphere (TOA) reflectance from GOCI multispectral bands and AErosol RObotic NETwork measurements, it showed high spatial and temporal generalization performance owing to its deeper and more complicated network structure. Hourly GOCI AOD data obtained using a deep learning approach with high accuracy are expected to be useful for the quantification of aerosol contents and monitoring of diurnal variations in the AOD.

To meet the growing demands of staple crops with a strategy to develop amicable strategic measures that support efficient North Korean relief policies, it is a desirable task to accurately simulate the yield of paddy (Oryza sativa), an important Asian food commodity. We aim to address this with a grid-based crop simulation model integrated with satellite imagery that enables us to monitor the crop productivity of North Korea. Vegetation Indices (VIs), solar insolation, and air temperature data are thus obtained from the Communication Ocean and Meteorological Satellite (COMS), including the reanalysis data of the Korea Local Analysis and Prediction System (KLAPS). Paddy productivities for North Korea are projected based on the bidirectional reflectance distribution function-adjusted VIs and the solar insolation using the grid GRAMI-rice model. The model is calibrated on a 500-m grid paddy field in Cheorwon, and the model simulation performance accuracy is verified for Cheorwon and Paju, located at the borders of North Korea using four years of data from 2011 to 2014. Our results show that the paddy yields are reproduced reasonably accurately within a statistically significant range of accuracy, in comparison with observation data in Cheorwon (p = 0.183), Paju (p = 0.075), and NK (p = 0.101) according to a statistical t-test procedure. We advocate that incorporating a crop model with satellite images for crop yield simulations can be utilised as a reliable estimation technique for the monitoring of crop productivity, particularly in unapproachable, data-sparse regions not only in North Korea, but globally, where estimations of paddy productivity can assist in planning of agricultural activities that support regionally amicable food security strategies.

Monitoring crop conditions and forecasting crop yields are both important for assessing crop production and for determining appropriate agricultural management practices; however, remote sensing is limited by the resolution, timing, and coverage of satellite images, and crop modeling is limited in its application at regional scales. To resolve these issues, the Gramineae (GRAMI)-rice model, which utilizes remote sensing data, was used in an effort to combine the complementary techniques of remote sensing and crop modeling. The model was then investigated for its capability to monitor canopy growth and estimate the grain yield of rice (Oryza sativa), at both the field and the regional scales, by using remote sensing images with high spatial resolution. The field scale investigation was performed using unmanned aerial vehicle (UAV) images, and the regional-scale investigation was performed using RapidEye satellite images. Simulated grain yields at the field scale were not significantly different (p = 0.45, p = 0.27, and p = 0.52) from the corresponding measured grain yields according to paired t-tests (α = 0.05). The model’s projections of grain yield at the regional scale represented the spatial grain yield variation of the corresponding field conditions to within ±1 standard deviation. Therefore, based on mapping the growth and grain yield of rice at both field and regional scales of interest within coverages of a UAV or the RapidEye satellite, our results demonstrate the applicability of the GRAMI-rice model to the monitoring and prediction of rice growth and grain yield at different spatial scales. In addition, the GRAMI-rice model is capable of reproducing seasonal variations in rice growth and grain yield at different spatial scales.

The Geostationary Ocean Color Imager (GOCI) of the Communication, Ocean, and Meteorological Satellite (COMS) increases the chance of acquiring images with greater clarity eight times a day and is equipped with spectral bands suitable for monitoring crop yield in the national scale with a spatial resolution of 500 m. The objectives of this study were to classify nationwide paddy fields and to project rice (Oryza sativa) yield and production using the grid-based GRAMI-rice model and GOCI satellite products over South Korea from 2011 to 2014. Solar insolation and temperatures were obtained from COMS and the Korea local analysis and prediction systems for model inputs, respectively. The paddy fields and transplanting dates were estimated by using Moderate Resolution Imaging Spectroradiometer (MODIS) reflectance and land cover products. The crop model was calibrated using observed yield data in 11 counties and was applied to 62 counties in South Korea. The overall accuracies of the estimated paddy fields using MODIS data ranged from 89.5% to 90.2%. The simulated rice yields statistically agreed with the observed yields with mean errors of −0.07 to +0.10 ton ha⁻¹, root-mean-square errors of 0.219 to 0.451 ton ha⁻¹, and Nash–Sutcliffe efficiencies of 0.241 to 0.733 in four years, respectively. According to paired t-tests (α = 0.05), the simulated and observed rice yields were not significantly different. These results demonstrate the possible development of a crop information delivery system that can classify land cover, simulate crop yield, and monitor regional crop production on a national scale.

Radiometric calibration of satellite imaging sensors should be performed periodically to account for the effect of sensor degradation in the space environment on image accuracy. In this study, we performed vicarious radiometric calibrations (relying on in situ data) of multispectral imaging sensors on the Korea multi-purpose satellite-3 and -3A (KOMPSAT-3 and -3A) to adjust the existing radiometric conversion coefficients according to time delay integration (TDI) adjustments and sensor degradation over time. The Second Simulation of a Satellite Signal in the Solar Spectrum (6S) radiative transfer model was used to obtain theoretical top of atmosphere radiances for both satellites. As input parameters for the 6S model, surface reflectance values of well-characterized pseudo-invariant tarps were measured using dual ASD FieldSpec^® 3 hyperspectral radiometers, and atmospheric conditions were measured using Microtops II^® Sunphotometer and Ozonometer. We updated the digital number (DN) of the radiance coefficients of the satellites; these had been used to calibrate the sensors during in-orbit test periods in 2013 and 2015. The coefficients of determination, R², values between observed DNs of the sensors, and simulated radiances for the tarps were more than 0.999. The calibration errors were approximately 5.7% based on manifested error sources. We expect that the updated coefficients will be an important reference for KOMPSAT-3 and -3A users.

This study proposes an improved approach for monitoring the spatial concentrations of hourly particulate matter less than 2.5 μm in diameter (PM_2.5) via a deep neural network (DNN) using geostationary ocean color imager (GOCI) images and unified model (UM) reanalysis data over the Korean Peninsula. The DNN performance was optimized to determine the appropriate training model structures, incorporating hyperparameter tuning, regularization, early stopping, and input and output variable normalization to prevent training dataset overfitting. Near-surface atmospheric information from the UM was also used as an input variable to spatially generalize the DNN model. The retrieved PM_2.5 from the DNN was compared with estimates from random forest, multiple linear regression, and the Community Multiscale Air Quality model. The DNN demonstrated the highest accuracy compared to that of the conventional methods for the hold-out validation (root mean square error (RMSE) = 7.042 μg/m³, mean bias error (MBE) = −0.340 μg/m³, and coefficient of determination (R²) = 0.698) and the cross-validation (RMSE = 9.166 μg/m³, MBE = 0.293 μg/m³, and R² = 0.49). Although the R² was low due to underestimated high PM_2.5 concentration patterns, the RMSE and MBE demonstrated reliable accuracy values (<10 μg/m³ and 1 μg/m³, respectively) for the hold-out validation and cross-validation.