China’s massive wave of urbanization may be threatened by land subsidence. Using a spaceborne synthetic aperture radar interferometry technique, we provided a systematic assessment of land subsidence in all of China’s major cities from 2015 to 2022. Of the examined urban lands, 45% are subsiding faster than 3 millimeters per year, and 16% are subsiding faster than 10 millimeters per year, affecting 29 and 7% of the urban population, respectively. The subsidence appears to be associated with a range of factors such as groundwater withdrawal and the weight of buildings. By 2120, 22 to 26% of China’s coastal lands will have a relative elevation lower than sea level, hosting 9 to 11% of the coastal population, because of the combined effect of city subsidence and sea-level rise. Our results underscore the necessity of enhancing protective measures to mitigate potential damages from subsidence.

The monitoring of tree species diversity is important for forest or wetland ecosystem service maintenance or resource management. Remote sensing is an efficient alternative to traditional field work to map tree species diversity over large areas. Previous studies have used light detection and ranging (LiDAR) and imaging spectroscopy (hyperspectral or multispectral remote sensing) for species richness prediction. The recent development of very high spatial resolution (VHR) RGB images has enabled detailed characterization of canopies and forest structures. In this study, we developed a three-step workflow for mapping tree species diversity, the aim of which was to increase knowledge of tree species diversity assessment using deep learning in a tropical wetland (Haizhu Wetland) in South China based on VHR-RGB images and LiDAR points. Firstly, individual trees were detected based on a canopy height model (CHM, derived from LiDAR points) by the local-maxima-based method in the FUSION software (Version 3.70, Seattle, USA). Then, tree species at the individual tree level were identified via a patch-based image input method, which cropped the RGB images into small patches (the individually detected trees) based on the tree apexes detected. Three different deep learning methods (i.e., AlexNet, VGG16, and ResNet50) were modified to classify the tree species, as they can make good use of the spatial context information. Finally, four diversity indices, namely, the Margalef richness index, the Shannon–Wiener diversity index, the Simpson diversity index, and the Pielou evenness index, were calculated from the fixed subset with a size of 30 × 30 m for assessment. In the classification phase, VGG16 had the best performance, with an overall accuracy of 73.25% for 18 tree species. Based on the classification results, mapping of tree species diversity showed reasonable agreement with field survey data (R²_Margalef = 0.4562, root-mean-square error RMSE_Margalef = 0.5629; R²_{Shannon–Wiener} = 0.7948, RMSE_{Shannon–Wiener} = 0.7202; R²_Simpson = 0.7907, RMSE_Simpson = 0.1038; and R²_Pielou = 0.5875, RMSE_Pielou = 0.3053). While challenges remain for individual tree detection and species classification, the deep-learning-based solution shows potential for mapping tree species diversity.

Lakes play a key role in the global water cycle, providing essential water resources and ecosystem services for humans and wildlife. Quantifying long-term changes in lake volume at a global scale is therefore important to the sustainability of humanity and natural ecosystems. Yet, such an estimate is still unavailable because, unlike lake area, lake volume is three-dimensional, challenging to be estimated consistently across space and time. Here, taking advantage of recent advances in remote sensing technology, especially NASA’s ICESat-2 satellite laser altimeter launched in 2018, we generated monthly volume series from 2003 to 2020 for 9065 lakes worldwide with an area ≥ 10 km². We found that the total volume of the 9065 lakes increased by 597 km³ (90% confidence interval 239–2618 km³). Validation against in situ measurements showed a correlation coefficient of 0.98, an RMSE (i.e., root mean square error) of 0.57 km³ and a normalized RMSE of 2.6%. In addition, 6753 (74.5%) of the lakes showed an increasing trend in lake volume and were spatially clustered into nine hot spots, most of which are located in sparsely populated high latitudes and the Tibetan Plateau; 2323 (25.5%) of the lakes showed a decreasing trend in lake volume and were clustered into six hot spots—most located in the world’s arid/semi-arid regions where lakes are scarce, but population density is high. Our results uncovered, from a three-dimensional volumetric perspective, spatially uneven lake changes that aggravate the conflict between human demands and lake resources. The situation is likely to intensify given projected higher temperatures in glacier-covered regions and drier climates in arid/semi-arid areas. The 15 hot spots could serve as a blueprint for prioritizing future lake research and conservation efforts.

Sensors onboard satellite platforms with short revisiting periods acquire frequent earth observation data. One limitation to the utility of satellite-based data is missing information in the time series of images due to cloud contamination and sensor malfunction. Most studies on gap-filling and cloud removal process individual images, and existing multi-temporal image restoration methods still have problems in dealing with images that have large areas with frequent cloud contamination. Considering these issues, we proposed a deep learning-based method named content-sequence-texture generation (CSTG) network to generate gap-filled time series of images. The method uses deep neural networks to restore remote sensing images with missing information by accounting for image contents, textures and temporal sequences. We designed a content generation network to preliminarily fill in the missing parts and a sequence-texture generation network to optimize the gap-filling outputs. We used time series of Moderate-resolution Imaging Spectroradiometer (MODIS) data in different regions, which include various surface characteristics in North America, Europe and Asia to train and test the proposed model. Compared to the reference images, the CSTG achieved structural similarity (SSIM) of 0.953 and mean absolute errors (MAE) of 0.016 on average for the restored time series of images in artificial experiments. The developed method could restore time series of images with detailed texture and generally performed better than the other comparative methods, especially with large or overlapped missing areas in time series. Our study provides an available method to gap-fill time series of remote sensing images and highlights the power of the deep learning methods in reconstructing remote sensing images.

Automatic classification of light detection and ranging (LiDAR) data in urban areas is of great importance for many applications such as generating three-dimensional (3D) building models and monitoring power lines. Traditional supervised classification methods require training samples of all classes to construct a reliable classifier. However, complete training samples are normally hard and costly to collect, and a common circumstance is that only training samples for a class of interest are available, in which traditional supervised classification methods may be inappropriate. In this study, we investigated the possibility of using a novel one-class classification algorithm, i.e., the presence and background learning (PBL) algorithm, to classify LiDAR data in an urban scenario. The results demonstrated that the PBL algorithm implemented by back propagation (BP) neural network (PBL-BP) could effectively classify a single class (e.g., building, tree, terrain, power line, and others) from airborne LiDAR point cloud with very high accuracy. The mean F-score for all of the classes from the PBL-BP classification results was 0.94, which was higher than those from one-class support vector machine (SVM), biased SVM, and maximum entropy methods (0.68, 0.82 and 0.93, respectively). Moreover, the PBL-BP algorithm yielded a comparable overall accuracy to the multi-class SVM method. Therefore, this method is very promising in the classification of the LiDAR point cloud.