Urban vegetation is a critical component of the terrestrial ecosystem, and accurately calculating its carbon storage is essential, particularly in the context of carbon accounting. Net Primary Productivity (NPP) is a key indicator of the surface carbon cycle, reflecting the health and robustness of terrestrial ecosystems and serving as a fundamental measure for carbon storage. However, most existing studies rely on medium- to low-resolution data and annual-scale temporal resolution for NPP estimation. These approaches often fail to capture the dynamic nature of vegetation growth, especially during the growing season, due to limitations like cloud cover, snow, transmission errors, and data coverage gaps. The challenge of capturing vegetation growth dynamics is particularly pronounced in urban areas, where surface heterogeneity and vegetation fragmentation complicate accurate NPP estimation. In this study, we focus on the Pearl River Delta region and address these data limitations by reconstructing high-frequency time series Normalized Difference Vegetation Index (NDVI) using remote sensing image fusion based on MODIS and Landsat. NPP estimation is conducted using the improved Carnegie-Ames-Stanford Approach model, which utilizes detailed land cover classification, time series NDVI, and climate data to achieve high spatial and temporal resolution results. Our results demonstrate a strong correlation between the NPP estimates derived from NDVI reconstruction data and the MOD17A3-NPP product. According to research findings from 2017 to 2020, the total vegetation NPP in the PRD decreased from 1731.9 GgC/y to 1500.2 GgC/y. The 30 m monthly NPP time series provides a more accurate reflection of vegetation growth dynamics. These high-resolution NPP products are crucial for precise carbon storage estimation, enhancing our understanding of urban ecosystems, and offering a detailed metric for urban carbon accounting.

Multi-modal remote sensing image (MRSI) matching has always been a challenging task. Traditional image matching methods often fail to obtain satisfactory results in most cases due to temporal differences, complex geometric distortions, and non-linear radiometric differences (NRDs). The key to addressing MRSI matching lies in mitigating NRDs to achieve robust extraction and description of features. This paper proposes a multiscale phase-orientation guided feature transform (MS-POFT) for multi-modal image matching. Two novel strategies are investigated and integrated into MS-POFT to improve the matching performance. A phase-structured adaptive detection is designed by the complementation of phase stretching transform and adaptive sliding windows, which ensures stable feature point extraction across different scales. Then, a new feature descriptor suitable for multi-modal images, called MS-PGLOH, is constructed based on phase and gradient principal direction in multiscale space. We performed comparison experiments on various multimodal datasets from remote sensing, natural sceneries, night surveillance, medical and temporal changes. Our experimental results both in qualitative and quantitative ways show that our proposed MS-POFT outperforms other comparison methods. MS-POFT successfully matched all given image pairs, achieving satisfactory results in terms of the number of correct matches (NCM), proportion of corrections ratio (PCR), and a reduced root-mean-square error (RMSE) of approximately 1.36.

As China’s rapid urban development exerts significant pressure on the ecological environment, accurate quantitative analysis of the relationship between urban and eco-environment is crucial for sustainable development. Current research has not yet explored the spatial and temporal patterns of the coupling relationship between urban and ecological systems at the county scale. To address this gap, this study developed a multi-scale analysis framework for ecological-urban resilience based on geographically weighted principal component analysis (GWPCA). Based on data from 1068 counties across the Yangtze River Economic Belt (YREB), 21 evaluation indicators were selected to explore the spatiotemporal patterns in the coupling coordination degree (CCD) between ecological resilience (ER) and urban resilience (UR) at different scales. The results indicated the following: (1) the ER of YREB exhibited a spatial pattern of “high in the southeast and low in the northwest”, while the UR exhibited a gradual and steady growth trend; (2) the CCD of YREB underwent three stages, with the development stage progressing 1.58 times faster than the starting stage, and the CCD peaked at 0.663 in 2020; (3) the CCD showed an upward trend from the upper reaches of the Yangtze River (URYR) to the lower reaches of the Yangtze River (LRYR), corresponding to improvements in urban development levels and the eco-environment quality. Under the “Ecological Priority and Intensive Development” strategy, the ER, UR, and CCD of YREB increased continuously from 2000 to 2020. Furthermore, targeted policies should be developed in response to the substantial CCD gap between the eastern and western regions and the decline in CCD of certain counties. The research findings will provide valuable scientific data and recommendations for environmental protection, ecological restoration, and innovation-driven regional development in the YREB.

The deep learning-based building extraction methods produce different feature maps at different stages of the network, which contain different information features. The detailed information of the feature maps decreases along the depth of the network, and insufficiently detailed information results in limited accuracy. However, existing methods are incapable of making full use of low-level feature maps with rich details. To overcome these shortcomings, we proposed a Cross-stage Features Fusion Network (CFF-Net) for building extraction from remote sensing images. In the CFF-Net, we innovatively proposed a Cross-stage Features Fusion (CFF) module that fuses different features generated at different stages. And we used the attention mechanism to make the network more focused on important information at different scales. To further improve the accuracy of building extraction, we designed the Prediction Enhancement (PE) module, where the last convolutional layer and the feature map generated in the intermediate stage are used for prediction at the same time to enhance the final result. To evaluate the effectiveness of the proposed network, we conduct quantitative and qualitative experiments on the two publicly available datasets, i.e. the Inria dataset and the WHU datasets. CFF-Net outperformed other state-of-the-art algorithms on the two datasets in IoU and F1 metrics. The efficiency analysis reveals that the proposed CFF-Net achieves a great balance between building extraction performance and complexity/efficiency, with faster convergence and higher robustness.