A study published in Computational Urban Science reveals significant spatial differences in how heat waves affect the emotional well-being of residents in Beijing. By analyzing multi-source geographic big data, researchers identified that emotional responses to extreme heat are not uniform across the city. The study, led by Yali Wei and colleagues, utilized social media data from Sina Weibo to extract fine-grained emotional information. This approach allowed for a detailed examination of how different urban environments influence specific emotions during heat wave events.

The research team employed a dictionary-based sentiment analysis method to categorize emotions into seven distinct types: good, happy, surprise, anger, disgust, fear, and sadness. They processed 11.62 million Weibo records from 2019, which included geographic coordinates and timestamps. This large dataset enabled the researchers to map emotional expressions at the subdistrict level across 327 administrative units in Beijing. The use of social media data provided a timely and large-scale alternative to traditional questionnaire surveys, which often struggle to capture real-time emotional fluctuations.

Heat waves in Beijing were defined as periods where the daily maximum temperature exceeded 35 °C for three or more consecutive days. Four such events were identified in 2019, totaling 20 days. The study integrated remote sensing imagery, including Sentinel-2 and DMSP-OLS data, with point-of-interest (POI) data to characterize the urban built environment. Key indicators included the normalized difference vegetation index (NDVI), nighttime light intensity (NTL), normalized difference built-up index (NDBI), and POI entropy density. These variables reflect urban ecological buffering capacity, thermal exposure, and socioeconomic activity intensity.

The results showed that residents in suburban areas tended to express emotions such as “surprise” and “happy” during heat waves. In contrast, residents in central urban areas more frequently expressed “good” and “sadness.” This spatial heterogeneity highlights the complex relationship between urban form and emotional response. The study found that positive emotions like “happy” and “good” were more prominent in the southeastern districts, such as Daxing and Tongzhou. Negative emotions like “sadness” and “disgust” were concentrated in the city center and its surrounding areas.

To investigate the causal factors behind these emotional patterns, the researchers used the Geographical Convergent Cross Mapping (GCCM) method. GCCM is an advanced spatial causal analysis approach that identifies nonlinear causal relationships within complex systems. Unlike traditional regression models, GCCM accounts for spatial interdependencies and nonlinearity, providing a more robust framework for causal inference. The method involves reconstructing shadow manifolds and evaluating cross-mapping prediction skills to determine causal associations between built environment factors and emotional expressions.

The GCCM analysis revealed that emotions such as “happy,” “surprise,” “fear,” and “anger” had stronger causal associations with urban functional indicators. Specifically, NDVI and NTL showed significant causal effects on these emotions. NDVI, which reflects urban greenery, was linked to the mitigation of heat stress and the promotion of mental well-being. NTL, representing urbanization and socioeconomic activity, was associated with higher psychological burdens and negative emotional responses. The study found that “good” was strongly associated with both NDVI and NTL, suggesting that green environments and economic vitality enhance short-term sensory pleasure and aesthetic experiences.

In contrast, POI entropy density and NDBI exhibited relatively weak causal relationships with the seven emotions. POI entropy density measures urban functional mix, with higher values indicating greater functional diversity and service accessibility. NDBI measures the proportion of impervious surfaces, which can enhance urban heat island effects. The lack of significant causal links for these variables suggests that they may have less direct influence on emotional expressions during heat waves compared to vegetation and nighttime light intensity.

The study also noted that “anger” and “surprise” were expressed less frequently in Beijing compared to Western contexts. This difference may be attributed to cultural norms that regulate emotional expression on social media. The researchers emphasized that fine-grained sentiment analysis overcomes the limitations of binary positive/negative classifications, allowing for a more nuanced understanding of emotional responses. By identifying specific emotions, the study provides insights into the differentiated impacts of heat waves on public health.

Limitations of the study include the potential representativeness bias of social media data, as Weibo users are predominantly young. This may limit the coverage of older adults and children, who might be more sensitive to extreme weather. Additionally, data sparsity in remote areas could lead to systematic biases in the spatial distribution of emotions. The study acknowledges that personal factors, such as occupation and income, were not included due to privacy constraints. Future research could incorporate questionnaire survey data to supplement the findings and explore the causal relationships on a national scale.

The findings of this study offer valuable insights for urban planning and public health strategies. By identifying the spatial heterogeneities in emotional responses to heat waves, the research supports the development of targeted and differentiated interventions. Urban planners can use these insights to optimize the built environment, enhancing green spaces and managing urban heat island effects to promote emotional well-being. The study underscores the importance of integrating multi-source geographic big data in understanding the complex interactions between climate, urban form, and human health.

To identify extreme heat episodes, researchers retrieved meteorological records from the China Meteorological Data Service Center, defining a heatwave as three or more consecutive days with maximum temperatures exceeding 35 degrees Celsius. This standard identified four heatwaves in 2019: May 22 to 25, June 23 to 26, July 1 to 4, and July 21 to 28, totaling twenty days. Administrative boundaries at 1:4,000,000 scale from the National Earth System Science Data Center partitioned Beijing into 327 subdistricts. Environmental variables were processed in ArcGIS to compute subdistrict averages. The vegetation and built-up indices were obtained from Google Earth Engine, while nighttime lights came from 500-meter monthly composites from the Earth Observation Group. Land-use diversity was evaluated across eleven categories using Amap data to compute Shannon entropy.

The linguistic framework utilized the DUTSO sentiment lexicon developed by Dalian University of Technology, based on Paul Ekman’s emotion theory and adapted to Chinese semantic characteristics. The lexicon contains seven primary categories, twenty-one subcategories, and 27,466 sentiment terms. Text from Sina Weibo underwent cleaning to remove duplicate and empty posts, followed by Jieba segmentation and stop-word removal. The algorithm matched emotional terms while incorporating negation words, which reversed polarity by multiplying weights by negative one, and degree adverbs that adjusted intensity. To evaluate accuracy, researchers randomly annotated 2,000 posts, constructing a two-by-one polarity matrix and a two-by-seven emotion matrix. Calculating precision, recall, and F1 scores confirmed strong model performance across polarity and specific emotional classes.

Spatial statistical analysis revealed distinct geographic clustering in emotional expression during heatwaves. The Global Moran’s I index for emotional post ratios across subdistricts reached 0.031 with a Z-score of 3.018, confirming significant positive spatial autocorrelation at p < 0.05. The proportion of emotional posts differed substantially across Beijing. Suburban subdistricts exhibited higher emotional expression ratios, peaking at 0.700 in Han Village within Fangshan District, whereas inner-city areas within the Second Ring Road showed lower proportions, such as 0.391 in Shichahai subdistrict. Anselin Local Moran’s I analysis showed that southwestern central subdistricts formed positive emotional clusters. Conversely, several near-suburban subdistricts in Fangshan recorded moderate sentiment averages and two negative values, generating a localized low-value emotional cluster.

District-level measurements documented detailed differences across specific negative emotions. Disgust occurred more frequently than sadness, concentrating in near-suburban and central districts. Disgust values were elevated in Tongzhou at 0.181, Shunyi at 0.178, Fengtai at 0.177, Daxing at 0.169, Changping at 0.169, Haidian at 0.168, and Chaoyang at 0.166. Sadness reached its highest levels in Fengtai at 0.083, Fangshan at 0.080, Tongzhou at 0.079, Haidian at 0.079, and Chaoyang at 0.074. Fear appeared at moderate frequencies across central districts, including Haidian at 0.030, Dongcheng at 0.029, and Xicheng at 0.029. Anger remained the least prevalent emotion, appearing mostly between the Fourth and Sixth Ring Roads, with highest rates in Fangshan at 0.009 and Yanqing at 0.008, whereas most Mentougou subdistricts recorded zero anger.

To infer nonlinear causal interactions, researchers implemented Geographical Convergent Cross Mapping in R using the stats and pcalg packages. Rather than relying on temporal lags, spatial GCCM reconstructs shadow manifolds using spatial lags between an administrative unit and its geographic neighbors. Causal intensity is assessed through cross-mapping prediction skill, measured by the Pearson correlation coefficient rho between observed values and cross-mapped estimates across expanding library sizes. Convergence of rho establishes a reliable causal relationship, evaluated at p < 0.05 with 95 percent confidence intervals. The models identified significant unidirectional causal relationships linking vegetation and nighttime light intensity to sadness, surprise, and fear. Conversely, point-of-interest entropy and built-up index values showed weak prediction skills that failed to achieve statistical significance.

Lexical analysis clarified the mechanisms linking built environment indicators to distinct positive emotions. In the DUTSO system, terms classified under happy, such as joyful or at peace, reflect cumulative psychological states tied to long-term stability. Conversely, vocabulary under good, including beautiful, handsome, or charming, represents immediate evaluative judgments responsive to aesthetic and visual stimuli. This distinction explains why vegetation and nighttime lighting demonstrated stronger causal influence on good by enhancing momentary sensory experiences. Methodologically, the study gathered all Weibo posts published during defined heatwave periods rather than filtering exclusively for high-temperature keywords, capturing direct and indirect expressions of heat stress. The authors noted that this sampling approach could capture unrelated chatter, suggesting that subsequent studies incorporate topic modeling and large language models to refine sentiment classification.

Subject of Research: Spatial heterogeneities of heat wave impacts on emotions and their causal factors in Beijing

Article Title: Spatial heterogeneities of heat waves’ impact on emotions and their causal factors based on multi-source geographic big data-a case study in Beijing city

Article References: Wei, Y., Wang, J., Zhu, Y., Yuan, Y., Li, H., Yu, J., & Meng, B. (2026). Spatial heterogeneities of heat waves’ impact on emotions and their causal factors based on multi-source geographic big data-a case study in Beijing city. Computational Urban Science, 6(1), Article 40. https://doi.org/10.1007/s43762-026-00272-7

Image Credits: AI Generated

DOI: 10.1007/s43762-026-00272-7

Keywords: heat waves, emotional well-being, Beijing, urban environment, social media analysis, GCCM method, spatial heterogeneity, public health, urban planning, sentiment analysis, Spatial, heterogeneities