The Pearl River Delta in South China is subject to severe eutrophication, which is significantly exacerbated by the total nitrogen (TN). Remote sensing technology with large-scale synchronous observations in the Pearl River Delta can effectively monitor organic pollution. Statistical methods based on remote sensing images have been widely used in water quality parameter retrieval for inland rivers, reservoirs, and lakes, but have seldom been applied in the Pearl River Delta. TN is also a non-optically active substance, so it is difficult to retrieve TN through analysis methods. This study retrieves the concentration of total nitrogen (TN) based on Landsat8 images of the Pearl River Delta using a statistical method. The stepwise regression function is built by analyzing the TN concentration and the single-band, two-band, and three-band spectral information groups measured by an ASD FieldSpec3 spectrometer. The retrieval results show that the proposed method performs well with a small mean absolute error (MAE) (0.36 mg/L for TN) and high agreement (R² = 0.61 for TN) between the in situ data and the retrieval concentration. The results demonstrate that the concentration of TN in the east of the Pearl River Delta was higher than in the west. Dachan Bay and Shenzhen Bay had the highest TN concentrations, which were around 3.02 mg/L and 3.67 mg/L. The 750–850 nm band could be an important reference for further exploring the spectral characteristics and retrieval of TN. The retrieval method in this study is easy to implement and convenient for local TN distribution capture, which can provide a timely reference for daily water quality supervision and management in the Pearl River Delta.

This work proposes a new sensitivity analysis model, referred to as the D-δ(ε) model, for the remote sensing retrieval of heavy metals in bodies of water. By defining the reflectance ratio function (δ(ε)), we deduce the mathematical relationships between the heavy metal concentration sequences (D_i) that can be effectively used for remote sensing retrievals and the radiometric resolution (ε) of the remote sensing instrument. Then, as a function of wavelength, we obtain the curve of the lower limit of the heavy metal concentrations in water that can be retrieved by remote sensing. To demonstrate the advantages of this model, we take two compounds, copper sulphate (CuSO₄) and cadmium sulphide (CdS), as examples to discuss the remote sensing sensitivity of different wavelengths when retrievals are performed using the Chinese HJ-1A's hyperspectral imager (HSI). The results showed that the lowest detectable concentration of CuSO₄ in the wavelength range of 460.04-496 nm (corresponding to bands 1-17 of the HSI image) can be below 0.15 mg/L, while the concentration of CdS can be lower than 0.001 mg/L in the separate ranges of 460.04-493.59 nm (bands 1-16) and 526.885-594.79 nm (bands 29-51). This model clearly demonstrates the mathematical relationship obeyed by "D-ε". Additionally, this model can not only calculate the retrieval concentration sequences at any observation wavelength but also intuitively provide the curve of the lower concentration limit for heavy metal retrievals. This work provides a theoretical basis for the selection of the most sensitive bands for remote sensing retrieval using hyperspectral images in the future.

The variation of sea surface temperature (SST) can change the backscatter coefficient measured by a scatterometer, resulting in a decrease in the accuracy of the sea surface wind measurement. This study proposed a new approach to correct the effect of SST on the backscatter coefficient. The method focuses on the Ku-band scatterometer HY-2A SCAT, which is more sensitive to SST than C-band scatterometers, can improve the wind measurement accuracy of the scatterometer without relying on reconstructed geophysical model function (GMF), and is more suitable for operational scatterometers. Through comparisons to WindSat wind data, we found that the Ku-band scatterometer HY-2A SCAT wind speeds are systemically lower under low SST and higher under high SST conditions. We trained a neural network model called the temperature neural network (TNNW) using HY-2A data and WindSat data. TNNW-corrected backscatter coefficients retrieved wind speed with a small systematic deviation from WindSat wind speed. In addition, we also carried out a validation of HY-2A wind and TNNW wind using European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis data as a reference, and the results showed that the retrieved TNNW-corrected backscatter coefficient wind speed is more consistent with ECMWF wind speed, indicating that the method is effective in correcting SST impact on HY-2A scatterometer measurements.

Numerous machine learning models have been widely used for the spatial prediction of PM_2.5 mass concentrations in the field of remote sensing, but most studies rely on single models, limiting their ability to capture complex nonlinear relationships. Furthermore, traditional Aerosol Optical Depth (AOD) methods suffer from extensive missing values due to algorithmic limitations, hindering daily PM_2.5 mass concentration retrieval. This study first developed a hybrid random forest and extreme gradient boosting model (RF-XGBoost) to overcome single-model accuracy constraints. Subsequently, Top-of-Atmosphere (TOA) reflectance replaced conventional AOD as the hybrid model’s input. Finally, we integrated four-year (2020–2023) TOA reflectance, normalized difference vegetation index (NDVI) data, meteorological data, digital elevation model (DEM) data, and day-of-year data to develop a high-precision hybrid model specifically optimized for Jiangxi Province. The simulation results demonstrated that the hybrid RF-XGBoost model (test-R² = 0.82, RMSE = 7.25 μg/m³, MAE = 4.90 μg/m³) outperformed the single Random Forest Model by 25% and 26% in terms of the root mean square error (RMSE) and mean absolute error (MAE), respectively. The high predictive accuracy of our method confirms its effectiveness in generating reliable PM_2.5 estimates. The resulting four-year dataset also successfully delineated the characteristic seasonal PM_2.5 pattern in the region, with the highest levels in winter and the lowest in summer, alongside a clear decreasing annual trend, signifying gradual atmospheric improvement.