Urban green spaces (UGSs) play a critical role in human thermal comfort, energy consumption and urban ecology. Although the heat mitigation capability of UGSs has been frequently reported, many of the current understandings are based on short-term observations, and the long-term temporal dynamics of UGS cooling effects are still lacking. This gap may cause over- or underestimation and largely ignores how the cooling effects change with climate change and urban growth. Accordingly, we used Landsat-based time series data to analyze the changes in permanent UGS greenness, surface-cooling effects and their biophysical responses in Beijing in the past 40 years (1984–2020). The results demonstrate segmented changes in UGS surface cooling that were mainly linked to the responses of canopy transpiration and albedo to vegetation conditions. During a rapid greening of UGSs in the recent two decades, transpiration cooling dominated albedo-induced warming to provide a discernable cooling enhancement. In addition, such enhancement showed seasonal differences ranging from less than 1 °C to more than 2 °C, and the most evident enhancement occurred on summer days (~2.4 °C) when vegetation is most needed to provide cooling. The highlighted dynamics of UGSs help urban planners better balance the maintenance costs and the environmental gains for UGS management.

During freeze–thaw periods, the exchange between shallow groundwater and soil water is unusually strong and bidirectional, which causes soil salinization and affects the accuracy of water resources assessment. The objectives of this study were to explore the laws of transformation between phreatic water and soil water through nine different groundwater table depths (GTDs) and three kinds of lithologies during three successive freeze–thaw periods using field lysimeters. The results showed that phreatic evaporation increased with smaller average soil particle sizes. The differences between phreatic evaporation and recharge to groundwater (DPR) and GTDs were well fitted by the semi-logarithmic model, and the regression coefficients A and B of the model were well fitted by the linear relationship with the average soil particle size. With the increase of soil particle size, the change of DPR decreased with the change rate of soil particle size. The extent of transformation between phreatic water and soil water decreased with the increase of soil particle size. During the whole freeze–thaw period, the negative value of DPR increased with an decrease in GTD. The groundwater depths of zero DPR (D-zero) of sandy loam, fine sand and sandy soil during the freeze–thaw periods were 2.79 m, 2.21 m and 2.12 m, respectively. This research is significant for the prevention of soil salinization disasters and the accurate assessment of water resources.

Influence of reservoir construction on surrounding vegetation cover

Accurate building segmentation has become critical in various fields such as urban management, urban planning, mapping, and navigation. With the increasing diversity in the number, size, and shape of buildings, convolutional neural networks have been used to segment and extract buildings from such images, resulting in increased efficiency and utilization of image features. We propose a building semantic segmentation method to improve the traditional Unet convolutional neural network by integrating attention mechanism and boundary detection. The attention mechanism module combines attention in the channel and spatial dimensions. The module captures image feature information in the channel dimension using a one-dimensional convolutional cross-channel method and automatically adjusts the cross-channel dimension using adaptive convolutional kernel size. Additionally, a weighted boundary loss function is designed to replace the traditional semantic segmentation cross-entropy loss to detect the boundary of a building. The loss function optimizes the extraction of building boundaries in backpropagation, ensuring the integrity of building boundary extraction in the shadow part. Experimental results show that the proposed model AMBDNet achieves high-performance metrics, including a recall rate of 0.9046, an IoU of 0.7797, and a pixel accuracy of 0.9140 on high-resolution remote sensing images, demonstrating its robustness and effectiveness in precise building segmentation. Experimental results further indicate that AMBDNet improves the single-class recall of buildings by 0.0322 and the single-class pixel accuracy by 0.0169 in the high-resolution remote sensing image recognition task.