China has undergone rapid urbanization and internal migration in the past few years, and its up-to-date gridded population datasets are essential for various applications. Existing datasets for China, however, suffer from either outdatedness or failure to incorporate data from the latest Seventh National Population Census of China, conducted in 2020. In this study, we develop a novel population downscaling approach that leverages stacking ensemble learning and big geospatial data to produce up-to-date population grids at a 100 m resolution for China using seventh census data at both county and town levels. The proposed approach employs stacking ensemble learning to integrate the strengths of random forest, XGBoost, and LightGBM through fusing their predictions in a training mechanism, and it delineates the inhabited areas from big geospatial data to enhance the gridded population estimation. Experimental results demonstrate that the proposed approach exhibits the best-fit performance compared to individual base models. Meanwhile, the out-of-sample town-level test set indicates that the estimated gridded population dataset (R²=0.8936) is more accurate than existing WorldPop (R²=0.7427) and LandScan (R²=0.7165) products for China in 2020. Furthermore, with the inhabited area enhancement, the spatial distribution of population grids is intuitively more reasonable than the two existing products. Hence, the proposed population downscaling approach provides a valuable option for producing gridded population datasets. The estimated 100 m gridded population dataset of China holds great significance for future applications, and it is publicly available at https://doi.org/10.6084/m9.figshare.24916140.v1 (Chen et al., 2024b).

In recent years, the rapid development of Earth observation technology has produced an increasing growth in remote sensing big data, posing serious challenges for effective and efficient processing and analysis. Meanwhile, there has been a massive rise in deep-learning-based algorithms for remote sensing tasks, providing a large opportunity for remote sensing big data. In this article, we initially summarize the features of remote sensing big data. Subsequently, following the pipeline of remote sensing tasks, a detailed and technical review is conducted to discuss how deep learning has been applied to the processing and analysis of remote sensing data, including geometric and radiometric processing, cloud masking, data fusion, object detection and extraction, land-use/cover classification, change detection and multitemporal analysis. Finally, we discussed technical challenges and concluded directions for future research in deep-learning-based applications for remote sensing big data.

Farmland parcel-based crop classification using satellite data plays an important role in precision agriculture. In this study, a deep-learning-based time-series analysis method employing optical images and synthetic aperture radar (SAR) data is presented for crop classification for cloudy and rainy regions. Central to this method is the spatial-temporal incorporation of high-resolution optical images and multi-temporal SAR data and deep-learning-based time-series analysis. First, a precise farmland parcel map is delineated from high-resolution optical images. Second, pre-processed SAR intensity images are overlaid onto the parcel map to construct time series of crop growth for each parcel. Third, a deep-learning-based (using the long short-term memory, LSTM, network) classifier is employed to learn time-series features of crops and to classify parcels to produce a final classification map. The method was applied to two datasets of high-resolution ZY-3 images and multi-temporal Sentinel-1A SAR data to classify crop types in Hunan and Guizhou of China. The classification results, with an 5.0% improvement in overall accuracy compared to those of traditional methods, illustrate the effectiveness of the proposed framework for parcel-based crop classification for southern China. A further analysis of the relationship between crop calendars and change patterns of time-series intensity indicates that the LSTM model could learn and extract useful features for time-series crop classification.

Water is a fundamental element in urban environments, and water body extraction is important for landscape and urban planning. Remote sensing has increasingly been used for water body extraction; however, in urban environments, this kind of approaches is challenging because of the significant within-class spectral variance in water areas and the presence of complex ground features. The objective of this study is to develop an automatic method that could improve water body extraction in urban environments from moderate spatial resolution satellite images. Central to our method is the combined use of multiscale extractions and spectral mixture analysis techniques in adaptive local regions. Specifically, we first calculate the NDWI image from experimental images for selecting water sample pixels. Second, on the basis of the selected water pixels, we apply an improved spectral mixture analysis technique on the experimental image to get water abundance of every pixel, and segment the abundance image to extract water bodies at the global scale. Third, in a similar manner, we iteratively conduct the water body extraction in multiscale local regions to refine the water bodies. Finally, the final result of water bodies is obtained when a stopping criterion is satisfied. We have implemented this method to produce water maps from an ALOS/AVNIR-2 image and a Terra/ASTER image covering urban areas. The experimental results illustrate that the proposed method has substantially outperformed two related methods that use the NDWI-based thresholding and the SVM classification for the entire image.

Reconstructing the missing data for cloud/shadow-covered optical satellite images has great significance for enhancing the data availability and multi-temporal analysis. In this study, we proposed a deep-learning-based method for cloud/shadow-covered missing data reconstruction for time-series Landsat images. Central to this method is the combined use of autoencoder, long-short-term memory (AE-LSTM)-based similar pixel clustering and for backward LSTM-based time-series prediction. First, manually delineated cloud/shadow-covered masks were overlaid onto multi-temporal satellite images to produce pixel-wise time-series data with masking values. Second, these pixel-wise time series were clustered by an AE-LSTM-based unsupervised method into multiple clusters, for searching similar pixels. Third, for each cluster of target images, a for-backward-LSTM-based model was established to restore missing values in time series data. Finally, reconstructed data were merged with cloud-free (image) regions to produce cloud-free time-series images. The proposed method was applied onto three datasets of multi-temporal Landsat-8 OLI images to restore cloud/shadow-covered images. The reconstruction results, showing an improvement of greater than 10% in normalized mean-square error compared to the state-of-the-art methods, demonstrated the effectiveness of the proposed method in time-series reconstruction for satellite images.

Pixel-based and object-based classifications are two commonly used approaches in extracting land cover information from remote sensing images. However, they each have their own inherent merits and limitations. This study, therefore, proposes a new classification method through the integration of pixel-based and object-based classifications (IPOC). Firstly, it employs pixel-based soft classification to obtain the class proportions of pixels to characterize the land cover details from pixel-scale properties. Secondly, it adopts area-to-point kriging to explore the class spatial dependence between objects for each pixel from object-based soft classification results. Thirdly, the class proportions of pixels and the class spatial dependence of pixels are fused as the class occurrence of pixels. Last, a linear optimization model on objects is built to determine the optimal class label of pixels within each object. Two remote sensing images are used to evaluate the effectiveness of IPOC. The experimental results demonstrate that IPOC performs better than the traditional pixel-based hard classification and object-based hard classification methods. Specifically, the overall accuracy of IPOC is 7.64% higher than that of pixel-based hard classification and 4.64% greater than that of object-based hard classification in the first experiment, while the overall accuracy improvements in the second experiment are 3.59% and 3.42%, respectively. Meanwhile, IPOC produces less salt and pepper effect than the pixel-based hard classification method and generates more accurate land cover details and small patches than the object-based hard classification method.