Owing to the limitation of spatial resolution and spectral resolution, deep learning methods are rarely used for the classification of multispectral remote sensing images based on the real spectral dataset from multispectral remote sensing images. This study explores the application of a deep learning model to the spectral classification of multispectral remote sensing images. To address the problem of the large workload with respect to selecting training samples during classification by deep learning, first, linear spectral mixture analysis and the spectral index method were applied to extract the pixels of impervious surfaces, soil, vegetation, and water. Second, through the Euclidean distance threshold method, a spectral dataset of multispectral image pixels was established. Third, a deep learning classification model, ResNet-18, was constructed to classify Landsat 8 OLI images based on pixels’ real spectral information. According to the accuracy assessment, the results show that the overall accuracy of the classification results can reach 0.9436, and the kappa coefficient can reach 0.8808. This study proposes a method that allows for the more optimized establishment of the actual spectral dataset of ground objects, addresses the limitations of difficult sample selection in deep learning classification and of spectral similarity in traditional classification methods, and applies the deep learning method to the classification of multispectral remote sensing images based on a real spectral dataset.

Land surface characteristics, including soil type, terrain slope, and antecedent soil moisture, have significant impacts on surface runoff during heavy precipitation in highly urbanized areas. In this study, a Linear Spectral Mixture Analysis (LSMA) method is modified to extract high-precision impervious surface, vegetation, and soil fractions. In the modified LSMA method, the representative endmembers are first selected by combining a high-resolution image from Google Earth; the unmixing results of the LSMA are then post-processed to reduce errors of misclassification with Normalized Difference Built-up Index (NDBI) and Normalized Difference Vegetation Index (NDVI). The modified LSMA is applied to the Landsat 8 Operational Land Imager (OLI) image from 18 October 2015 of the main urban area of Guangzhou city. The experimental result indicates that the modified LSMA shows improved extraction performance compared with the conventional LSMA, as it can significantly reduce the bias and root-mean-square error (RMSE). The improved impervious surface, vegetation, and soil fractions are used to calculate the composite curve number (CN) for each pixel according to the Soil Conservation Service curve number (SCS-CN) model. The composite CN is then adjusted with regional data of the terrain slope and total 5-day antecedent precipitation. Finally, the surface runoff is simulated with the SCS-CN model by combining the adjusted CN and real precipitation data at 1 p.m., 4 May 2015.

For the large-area snow depth (SD) data sets with high spatial resolution in the Altay region of Northern Xinjiang, China, we present a deterministic ensemble Kalman filter (DEnKF)-albedo assimilation scheme that considers the common land model (CoLM) subgrid heterogeneity. In the albedo assimilation of DEnKF-albedo, the assimilated albedos over each subgrid tile are estimated with the MCD43C1 bidirectional reflectance distribution function (BRDF) parameters product and CoLM calculated solar zenith angle. The BRDF parameters are hypothesized to be consistent over all subgrid tiles within a specified grid. In the SCF assimilation of DEnKF-albedo, a DEnKF combining a snow density-based observation operator considers the effects of the CoLM subgrid heterogeneity and is employed to assimilate MODIS SCF to update SD states over all subgrid tiles. The MODIS SCF over a grid is compared with the area-weighted sum of model predicted SCF over all the subgrid tiles within the grid. The results are validated with in situ SD measurements and AMSR-E product. Compared with the simulations, the DEnKF-albedo scheme can reduce errors of SD simulations and accurately simulate the seasonal variability of SD. Furthermore, it can improve simulations of SD spatiotemporal distribution in the Altay region, which is more accurate and shows more detail than the AMSR-E product.