Suspended particulate matter (SPM) is one of the dominant water constituents in inland and coastal waters, and SPM concnetration (C_SPM) is a key parameter describing water quality. This study, using in-situ spectral and C_SPM measurements as well as Sentinel 2 Multispectral Imager (MSI) images, aimed to develop C_SPM retrieval models and further to estimate the C_SPM values of Poyang Lake, China. Sixty-eight in-situ hyperspectral measurements and relative spectral response function were applied to simulate Sentinel 2 MIS spectra. Thirty-four samples were used to calibrate and the left samples were used to validate C_SPM retrieval models, respectively. The developed models were then applied to two Sentinel 2 MSI images captured in wet and dry seasons, and the derived C_SPM values were compared with those derived from MODIS B1 (λ = 645 nm). Results showed that the Sentinel 2 MSI B4–B8b models achieved acceptable to high fitting accuracies, which explained 81–93% of the variation of C_SPM. The validation results also showed the reliability of these six models, and the estimated C_SPM explained 77–93% of the variation of measured C_SPM with the mean absolute percentage error (MAPE) ranging from 36.87% to 21.54%. Among those, a model based on B7 (λ = 783 nm) appeared to be the most accurate one. The Sentinel 2 MSI-derived C_SPM values were generally consistent in spatial distribution and magnitude with those derived from MODIS. The C_SPM derived from Sentinel 2 MSI B7 showed the highest consistency with MODIS on 15 August 2016, while the Sentinel 2 MSI B4 (λ = 665 nm) produced the highest consistency with MODIS on 2 April 2017. Overall, this study demonstrated the applicability of Sentinel 2 MSI for C_SPM retrieval in Poyang Lake, and the Sentinel 2 MSI B4 and B7 are recommended for low and high loadings of SPM, respectively.

Decision tree classification is one of the most efficient methods for obtaining land use/land cover (LULC) information from remotely sensed imageries. However, traditional decision tree classification methods cannot effectively eliminate the influence of mixed pixels. This study aimed to integrate pixel unmixing and decision tree to improve LULC classification by removing mixed pixel influence. The abundance and minimum noise fraction (MNF) results that were obtained from mixed pixel decomposition were added to decision tree multi-features using a three-dimensional (3D) Terrain model, which was created using an image fusion digital elevation model (DEM), to select training samples (ROIs), and improve ROI separability. A Landsat-8 OLI image of the Yunlong Reservoir Basin in Kunming was used to test this proposed method. Study results showed that the Kappa coefficient and the overall accuracy of integrated pixel unmixing and decision tree method increased by 0.093% and 10%, respectively, as compared with the original decision tree method. This proposed method could effectively eliminate the influence of mixed pixels and improve the accuracy in complex LULC classifications.

In constructing models for predicting soil organic matter (SOM) by using visible and near-infrared (vis–NIR) spectroscopy, the selection of representative calibration samples is decisive. Few researchers have studied the inclusion of spectral pretreatments in the sample selection strategy. We collected 108 soil samples and applied six commonly used spectral pretreatments to preprocess soil spectra, namely, Savitzky–Golay (SG) smoothing, first derivative (FD), logarithmic function log(1/R), mean centering (MC), standard normal variate (SNV), and multiplicative scatter correction (MSC). Then, the Kennard–Stone (KS) strategy was used to select calibration samples based on the pretreated spectra, and the size of the calibration set varied from 10 samples to 86 samples (80% of the total samples). These calibration sets were employed to construct partial least squares regression models (PLSR) to predict SOM, and the built models were validated by a set of 21 samples (20% of the total samples). The results showed that 64−78% of the calibration sets selected by the inclusion of pretreatment demonstrated significantly better performance of SOM estimation. The average improved residual predictive deviations (ΔRPD) were 0.06, 0.13, 0.19, and 0.13 for FD, log(1/R), MSC, and SNV, respectively. Thus, we concluded that spectral pretreatment improves the sample selection strategy, and the degree of its influence varies with the size of the calibration set and the type of pretreatment.

The size of phytoplankton not only influences its physiology, metabolic rates and marine food web, but also serves as an indicator of phytoplankton functional roles in ecological and biogeochemical processes. Therefore, some algorithms have been developed to infer the synoptic distribution of phytoplankton cell size, denoted as phytoplankton size classes (PSCs), in surface ocean waters, by the means of remotely sensed variables. This study, using the NASA bio-Optical Marine Algorithm Data set (NOMAD) high performance liquid chromatography (HPLC) database, and satellite match-ups, aimed to compare the effectiveness of modeling techniques, including partial least square (PLS), artificial neural networks (ANN), support vector machine (SVM) and random forests (RF), and feature selection techniques, including genetic algorithm (GA), successive projection algorithm (SPA) and recursive feature elimination based on support vector machine (SVM-RFE), for inferring PSCs from remote sensing data. Results showed that: (1) SVM-RFE worked better in selecting sensitive features; (2) RF performed better than PLS, ANN and SVM in calibrating PSCs retrieval models; (3) machine learning techniques produced better performance than the chlorophyll-a based three-component method; (4) sea surface temperature, wind stress, and spectral curvature derived from the remote sensing reflectance at 490, 510, and 555 nm were among the most sensitive features to PSCs; and (5) the combination of SVM-RFE feature selection techniques and random forests regression was recommended for inferring PSCs. This study demonstrated the effectiveness of machine learning techniques in selecting sensitive features and calibrating models for PSCs estimations with remote sensing.

The Guangdong–Hong Kong–Macau Greater Bay Area (GBA) of China is one of the largest bay areas in the world. However, the spatiotemporal characteristics and driving mechanisms of urban expansions in this region are poorly understood. Here we used the annual remote sensing images, Geographic Information System (GIS) techniques, and geographical detector method to characterize the spatiotemporal patterns of urban expansion in the GBA and investigate their driving factors during 1986–2017 on regional and city scales. The results showed that: the GBA experienced an unprecedented urban expansion over the past 32 years. The total urban area expanded from 652.74 km² to 8137.09 km² from 1986 to 2017 (approximately 13 times). The annual growth rate during 1986–2017 was 8.20% and the annual growth rate from 1986 to 1990 was the highest (16.89%). Guangzhou, Foshan, Dongguan, and Shenzhen experienced the highest urban expansion rate, with the annual increase of urban areas in 51.51, 45.54, 36.76, and 23.26 km² y⁻¹, respectively, during 1986–2017. Gross Domestic Product (GDP), income, road length, and population were the most important driving factors of the urban expansions in the GBA. We also found the driving factors of the urban expansions varied with spatial and temporal scales, suggesting the general understanding from the regional level may not reveal detailed urban dynamics. Detailed urban management and planning policies should be made considering the spatial and internal heterogeneity. These findings can enhance the comprehensive understanding of this bay area and help policymakers to promote sustainable development in the future.