The rapid development in deep learning and computer vision has introduced new opportunities and paradigms for building extraction from remote sensing images. In this paper, we propose a novel fully convolutional network (FCN), in which a spatial residual inception (SRI) module is proposed to capture and aggregate multi-scale contexts for semantic understanding by successively fusing multi-level features. The proposed SRI-Net is capable of accurately detecting large buildings that might be easily omitted while retaining global morphological characteristics and local details. On the other hand, to improve computational efficiency, depthwise separable convolutions and convolution factorization are introduced to significantly decrease the number of model parameters. The proposed model is evaluated on the Inria Aerial Image Labeling Dataset and the Wuhan University (WHU) Aerial Building Dataset. The experimental results show that the proposed methods exhibit significant improvements compared with several state-of-the-art FCNs, including SegNet, U-Net, RefineNet, and DeepLab v3+. The proposed model shows promising potential for building detection from remote sensing images on a large scale.

The surface urban heat island intensity (SUHII) when using nadir‐viewing radiometric and complete surface temperature (T_randT_c) was evaluated. The urban areas of the Kowloon peninsula and Hong Kong Island were selected and four daytime Landsat TM images and two nighttime ASTER images were collected to retrieveT_rand then modelT_cbased on a semi‐empirical model. SUHIIs were estimated using both the retrievedT_rand modelledT_c. High spatial resolution (HR) airborne thermal images (0.2 m) observed at 12:10 noon on Oct 24, 2017 were used to retrieveT_cdirectly. Results based on HR data and satellite data were consistent and indicated that the geometry of the built‐up space had a larger impact on SUHII when usingT_c(SUHIIc) thanT_r(SUHIIr). During daytime SUHIIc decreased while SUHIIr showed a very slight increase with building density. Both SUHIIc and SUHIIr decreased with higher building height but the rate of decrease of SUHIIc was higher than SUHIIr. Both SUHIIc and SUHIIr decreased with increasing building height variance and increased with increasing sky view factor (SVF). The rate of decrease with building height variance for SUHIIc was larger than SUHIIr. The rate of increase of SUHIIc with SVF was higher than SUHIIr. During nighttime, geometry effects on SUHIIc and SUHIIr were different from daytime. Both SUHIIc and SUHIIr increased with building density, while the rate of increase of SUHIIc with building density, as well as with building height, was much higher than SUHIIr. Both SUHIIc and SUHIIr decreased with SVF, but the rate of decrease of SUHIIc was higher than SUHIIr. Both SUHIIc and SUHIIr initially increased with building height variance and then remained approximately constant. We also evaluated the UHI intensity: SUHIIc was much closer to UHII than SUHIIr. Overall, building geometry had larger impact on SUHIIc than on SUHIIr, that is, SUHIIc is more representative of urban climate than SUHIIr.

The urban heat island effect poses a growing threat to human society, especially in densely populated and developed megacities. With the introduction of the Local Climate Zones (LCZ) framework, new perspectives and findings have been brought to urban heat island studies. This study investigated the cooling effect of vegetation and albedo on the surface urban heat island (SUHI) in the classification system of LCZ during different seasons, using three Chinese megacities as case study areas. Single-factor linear regression and Pearson’s correlation coefficient were applied to analyze the seasonal cooling effect of both albedo and the NDVI on the SUHI within different LCZs. The results show that (1) the variability of the SUHI is reflected in its dominance and intensity within certain LCZs in different cities and in the efficiency of cooling factors; (2) the cooling effect of vegetation is dominant in each season, and the cooling effect produced by albedo within specific seasons can be differentiated by LCZs. This study provides valuable information for the mitigation of the SUHI magnitude in specific regions and at specific times of the year.

Harmonization of satellite imagery provides a good opportunity for studying land surface temperature (LST) as well as the urban heat island effect. However, it is challenging to use the harmonized data for the study of LST due to the systematic bias between the LSTs from different satellites, which is highly influenced by sensor differences and the compatibility of LST retrieval algorithms. To fill this research gap, this study proposes the comparison of different LST images retrieved from various satellites that focus on Hong Kong, China, by applying diverse retrieval algorithms. LST images generated from Landsat-8 using the mono-window algorithm (MWA_L8) and split-window algorithm (SWA_L8) would be compared with the LST estimations from Sentinel-3 SLSTR and Himawari-8 using the split-window algorithm (SWA_S3 and SWA_H8). Intercomparison will also be performed through segregated groups of different land use classes both during the daytime and nighttime. Results indicate that there is a significant difference among the quantitative distribution of the LST data generated from these three satellites, with average bias of up to −1.80 K when SWA_H8 was compared with MWA_L8, despite having similar spatial patterns of the LST images. The findings also suggest that retrieval algorithms and the dominant land use class in the study area would affect the accuracy of image-fusion techniques. The results from the day and nighttime comparisons revealed that there is a significant difference between day and nighttime LSTs, with nighttime LSTs from different satellite sensors more consistent than the daytime LSTs. This emphasizes the need to incorporate as much night-time LST data as available when predicting or optimizing fine-scale LSTs in the nighttime, so as to minimize the bias. The framework designed by this study will serve as a guideline towards efficient spatial optimization and harmonized use of LSTs when utilizing different satellite images associated with an array of land covers and at different times of the day.

Characteristics of the urban environment (e.g., building density and road network) can influence the spread and transmission of coronavirus disease 2019 (COVID-19) within cities, especially in high-density high-rise built environments. Therefore, it is necessary to identify the key attributes of high-density high-rise built environments to enhance modelling of the spread of COVID-19. To this end, case studies for testing attributes for modelling development were performed in two densely populated Chinese cities with high-rise, high-density built environments (Hong Kong and Shanghai).The investigated urban environmental features included 2D and 3D urban morphological indices (e.g., sky view factor, floor area ratio, frontal area density, height to width ratio, and building coverage ratio), socioeconomic and demographic attributes (e.g., population), and public service points-of-interest (e.g., bus stations and clinics). The modelling effects of 3D urban morphological features on the infection rate are notable in urban communities. As the spatial scale becomes larger, the modelling effect of 2D built environment factors (e.g., building coverage ratio) on the infection rate becomes more notable. The influence of several key factors (e.g., the building coverage ratio and population density) at different scales can be considered when modelling the infection risk in urban communities. The findings of this study clarify how attributes of built environments can be applied to predict the spread of infectious diseases. This knowledge can be used to develop effective planning strategies to prevent and control epidemics and ensure healthy cities.

The spatial-temporal mismatch of water resources and socio-economic development in rapidly urbanized regions has been the focus of water resource management, and is one of the main limitations to sustainable development goals (SDGs). Guangdong Province is represented by a highly developed economy and society, and has been undergoing disproportionately rapid population growth during the past two decades. The uneven distribution and spatial mismatch of water resources have aggravated the contradictions between water supply demand. In this paper, we evaluate the matching degree of water resources and social economic elements, detect the spatial-temporal changing pattern of the matching degree, and reveal the changing mechanism using the combined methods of the Thiel index, the space–time Gini coefficient, and the Lorenz curve. The results show that (1) the temporal matching is relatively balanced and reasonable, while the spatial mismatch is prominent and deteriorating constantly, especially the connection between the amount of water and GDP; (2) the water volume pressure is mainly concentrated in the Peral River Delta and eastern and western Guangdong, while water consumption stress is relatively lower in northeastern Guangdong; and (3) the water volume inequality is dominated by an intercity difference and is primarily caused by regional differences. Based on the results, policy implications, such as the implementation of integrated water resource management plans, investment in the development of alternative water sources, as well as enhanced public education and the establishment of a water saving society, have been provided to alleviate the mismatch issue between water elements and socio-economic parameters, and to push the realization of water-related SDGs.