Automatically extracting buildings from remote sensing images with deep learning is of great significance to urban planning, disaster prevention, change detection, and other applications. Various deep learning models have been proposed to extract building information, showing both strengths and weaknesses in capturing the complex spectral and spatial characteristics of buildings in remote sensing images. To integrate the strengths of individual models and obtain fine-scale spatial and spectral building information, this study proposed a stacking ensemble deep learning model. First, an optimization method for the prediction results of the basic model is proposed based on fully connected conditional random fields (CRFs). On this basis, a stacking ensemble model (SENet) based on a sparse autoencoder integrating U-NET, SegNet, and FCN-8s models is proposed to combine the features of the optimized basic model prediction results. Utilizing several cities in Hebei Province, China as a case study, a building dataset containing attribute labels is established to assess the performance of the proposed model. The proposed SENet is compared with three individual models (U-NET, SegNet and FCN-8s), and the results show that the accuracy of SENet is 0.954, approximately 6.7%, 6.1%, and 9.8% higher than U-NET, SegNet, and FCN-8s models, respectively. The identification of building features, including colors, sizes, shapes, and shadows, is also evaluated, showing that the accuracy, recall, F1 score, and intersection over union (IoU) of the SENet model are higher than those of the three individual models. This suggests that the proposed ensemble model can effectively depict the different features of buildings and provides an alternative approach to building extraction with higher accuracy.

Roadsides are important urban public spaces where residents are in direct contact with the thermal environment. Understanding the effects of different vegetation types on the roadside thermal environment has been an important aspect of recent urban research. Although previous studies have shown that the thermal environment is related to the type and configuration of vegetation, remote sensing-based technology is not applicable for extracting different vegetation types at the roadside scale. The rapid development and usage of street view data provide a way to solve this problem, as street view data have a unique pedestrian perspective. In this study, we explored the effects of different roadside vegetation types on land surface temperatures (LSTs) using street view images. First, the grasses–shrubs–trees (GST) ratios were extracted from 19,596 street view images using semantic segmentation technology, while LST and normalized difference vegetation index (NDVI) values were extracted from Landsat-8 images using the radiation transfer equation algorithm. Second, the effects of different vegetation types on roadside LSTs were explored based on geographically weighted regression (GWR), and the different performances of the analyses using remotely sensed images and street view images were discussed. The results indicate that GST vegetation has different cooling effects in different spaces, with a fitting value of 0.835 determined using GWR. Among these spaces, the areas with a significant cooling effect provided by grass are mainly located in the core commercial area of Futian District, which is densely populated by people and vehicles; the areas with a significant cooling effect provided by shrubs are mainly located in the industrial park in the south, which has the highest industrial heat emissions; the areas with a significant cooling effect provided by trees are mainly located in the core area of Futian, which is densely populated by roads and buildings. These are also the areas with the most severe heat island effect in Futian. This study expands our understanding of the relationship between roadside vegetation and the urban thermal environment, and has scientific significance for the planning and guiding of urban thermal environment regulation.

Estimating building heights by generating disparity maps from multi-view satellite images through stereo matching remains challenging in urban research. However, disparity maps produced by both traditional stereo matching methods and deep learning methods often contain matching errors due to insufficient texture, multi-scale complex scenes, and building occlusions, leading to underestimation of building heights. To address this issue, we proposed a new multi-scale feature connected densely network (MFC-CNN) for stereo matching to generate high-quality disparity maps from GF-7 images. Then, a spatial forward intersection algorithm was used to further generate DSM from the disparity maps. Finally, a buffer analysis method considering roof type was adopted to combine DSM with building footprints for building height extraction. The disparity maps generated by MFC-CNN on the WHU-Stereo dataset achieved EPE and D1 of 1.208 and 7.49% respectively, outperforming SGM, StereoNet, and BGA-Net models. The accuracy of DSM data generated by the spatial forward intersection algorithm also reached the best, with RMSE, ME, SD, and MAE values of 1.58, 0.11, 1.62, and 1.36 respectively. Finally, building heights in Bao'an District of Shenzhen, Jinan, Beijing, and Shanghai were extracted. In the experimental results of a total of 94,671 samples in the four cities, the MAE and RMSE were 2.86 and 3.96, and the correlation coefficient R2 between the extracted building heights and the reference heights was 0.87. These results indicate that the building height extraction results of this study are highly consistent with the reference results, and also proves that this method exhibits strong stability and transferability in different regions.

Building function classification is fundamental for urban planning, management, and sustainability research. Conventional methods relying on socio-economic data or manual surveys are labor-intensive with limited spatial coverage. To address this gap, we propose a data-driven knowledge discovery framework to reveal morphological signatures of building functions using multi-scale 2D and 3D landscape indices derived from urban morphological data. First, we established a hierarchical index system with 29 indicators across single-building, inter-building, and landscape levels. Based on building footprints, height, and road network data of Shenzhen, China, we screened six key indices (Building Area, Perimeter-Area Fractal Dimension, Building Shape Coefficient, Building Height, Mean Patch Area, Floor Area Ratio) through correlation analysis, significance testing, and principal component analysis. We then constructed a quantitative mapping between the six indices and five building function types (residential, commercial, industrial, public, mixed-use) using K-means clustering. Results show that the six indices exhibit statistically significant differences across building function types, with the most distinct profiles between commercial and residential buildings. The established index–function mapping provides a knowledge-driven foundation for future building function inference, which will be used for classification, accuracy validation, and method comparison in a follow-up study. This framework offers a scalable and automatic approach to characterize urban form–function relationships using only morphological data.

Gridded evapotranspiration (ET) and gross primary productivity (GPP) products describe agricultural water and carbon relationships. Interpreting their fitted boundaries requires evidence of empirical support and reproducibility. We assessed an ET-conditioned vapour pressure deficit (VPD) boundary in Hetao croplands using PML-V2.2a and ERA5-Land for 2019–2024. The frozen sample represented 193 native PML cells and 21,491 cell–date ET/GPP units. Three procedures separated baseline reproduction, strict extraction from the unchanged surface and refitting on identical records. A local boundary persisted after inverse-multiplicity weighting and single-record reconstruction, covering 32.3–35.4% of qualified ET columns. Boundary occurrence was repeatable, but post-turn slope intervals from both ERA5 and PML cell×year block schemes included zero. The primary deep-layer wetness subgroup displacement was undefined on common support, and turning eligibility depended on the side-count rule in a post-hoc sensitivity. Historical MOD16/MYD16 ET–VPM GPP comparisons did not recover a comparable continuous branch, although periods, products and support changed together. These findings distinguish a recurring local product association from stable turning geometry. The assessment provides an auditable sequence for checking native source units, support, connectivity and uncertainty before interpreting ET–VPD–GPP boundaries for agricultural water management. Independent crop and water-supply observations are needed to test management interpretations.

Rising vapor pressure deficit (VPD) suppresses crop photosynthesis, but whether antecedent deep-layer wetness systematically moderates this constraint remains unresolved in irrigated agricultural regions. We assembled a monthly 0.1° panel for 2019–2024 across the North China Plain by integrating gross primary production (GPP), satellite solar-induced chlorophyll fluorescence (SIF), ERA5-Land hydroclimate, cropland fractions, and irrigation information. The primary sample contained 163,002 observations from 3,881 grid cells. Pixel and year–month fixed-effects models separated VPD into temperature-linked and humidity-deficit components and tested their interactions with a standardized antecedent deep-layer wetness index. Robustness was evaluated across cropland thresholds, irrigation formulations, extreme-VPD metrics, sample years, and placebo specifications. The temperature-linked VPD component was negatively associated with SIF (β = -0.123, 95% CI: -0.198 to -0.049) and GPP (β = -0.158, 95% CI: -0.226 to -0.090), whereas the humidity-deficit component was negative but less precisely estimated. Interactions of antecedent deep-layer wetness with both VPD components were small and uncertain for SIF and GPP, with all principal confidence intervals spanning zero. Cumulative extreme-VPD exposure was also negatively associated with GPP, but its moderation by antecedent deep-layer wetness remained uncertain. These patterns persisted across cropland and irrigation specifications. Equivalence tests did not establish negligible interactions; the evidence therefore indicates no reproducible regional moderation rather than proof of no effect. Antecedent deep-layer wetness may retain local agronomic value, but it did not systematically offset atmospheric-dryness constraints at the regional scale. Distinguishing persistent atmospheric demand from context-specific apparent buffering is important for evaluating climate resilience in intensively managed croplands.