Projects | Zhifeng Cheng

Bidirectional Impacts of Repeated Respiratory Infections on Mental Well-Beings

Sun, 01 Mar 2026 00:00:00 +0000

Abstract The negative effect of respiratory infections (Flu, COVID, etc.) on mental disorders has been widely demonstrated, but few studies link those evidences to repeated infections. Will the psychological trauma from reinfection diminish due to the previous infection experience, or will it accumulate? What roles do the number of infections, the time of infections, the time interval between the previous infection, and the transition between different diseases play in the process? This study collected over 1 million respiratory infections self-reported by 300,000 individuals on social media, together with 7 million users with no self-declaration as candidate controls. We tracked people’s expressed sentiment before and after different infection episodes by a quasi-experimental design.

How Multi-Hazard Early-Warning Systems Benefit Mobility Resilience to Tropical Cyclones

Wed, 07 Jan 2026 00:00:00 +0000

This project was led by Haiyan Liu and Jianghao Wang.

Abstract Multi-hazard early-warning systems (MHEWS) are critical for mitigating extreme weather impacts and enhancing disaster resilience. However, quantitative empirical evidence on how different types of early warnings individually and collectively trigger preventive actions and influence resilience remains limited. Here, using location-based human mobility data aggregated from over 1.1 billion mobile devices across Chinese cities, we quantified daily intracity human mobility responses to 21,126 early warning signals during 19 tropical cyclones (TCs) from 2021 to 2023. To represent disaster resilience under MHEWS protection, we developed a protected resilience index that integrates both the magnitude of mobility changes and recovery durations. We found that, compared with city-level TC warnings alone, combined multi-level, multi-hazard warnings resulted in a 52.4 % reduction in mobility during TC exposure days, thereby increasing avoided direct population exposure by around 57.1 %. Each additional warning type further shortened recovery times, collectively reducing recovery durations by at least 55.6 %, with larger effects observed for stronger TCs. Under MHEWS protection, protected resilience remained statistically similar between moderate-intensity TCs (34 kt and 50 kt) but declined significantly under severe (≥64 kt) conditions. Although absolute reductions in exposure were greater in high-frequency, coastal, and wealthier cities, relative improvements from MHEWS were more pronounced in less frequently affected, inland, and socioeconomically disadvantaged areas. Consequently, MHEWS significantly narrowed resilience disparities among cities facing equivalent hazard exposures. This study introduces a scalable, behaviour-based framework for quantifying early-warning effectiveness, highlighting the essential role of integrated multi-level and multi-hazard warnings in disaster preparedness across cities amid escalating climate risks.

Unravelling Dual Roles of Human Activity in Shaping Global Fire Regimes

Mon, 15 Dec 2025 00:00:00 +0000

This project is leading by Jianghao Wang and Zhiyi Zhang

Abstract Understanding the intertwined impacts of climate change and human activities on global fire regimes is critical, yet current evidence remains limited in distinguishing the specific roles of humans in driving or mitigating fire occurrences. This study addresses this gap by leveraging the unprecedented global reductions in human activity during the COVID-19 pandemic as a natural experiment. By comparing global fire occurrences in the first half of 2020 with their usual levels, we estimate the contributions of reduced human activity to shifts in fire patterns using a weekly dataset at 0.5° spatial resolution, derived from satellite observations of fires and complemented by meteorological data, pandemic-related restrictions on human mobility and activities. Our analysis reveals that reduced human activity led to an average global decline of 11.8% in fire incidence, with remarkable spatial heterogeneity indicating the dominant human roles in either igniting or controlling fires. Notably, fire occurrences in forests and areas with lower anthropogenic modifications experienced the most substantial declines, highlighting the varying sensitivity of ecosystems to human activity. These findings underscore the critical influence of human activity on fire regimes, suggesting that integrated fire management strategies incorporating frequency, intensity and characteristic of human behavior are essential for mitigating wildfire risks in vulnerable regions.

Connected and Contagious: How Social Connections, Mobility Networks, and Physical Contact Jointly Shape Infectious Disease Transmission

Wed, 03 Dec 2025 00:00:00 +0000

Abstract

Background

The interconnectedness of human society in this modern world can transform localised outbreaks into global pandemics, underscoring the pivotal roles of social, mobility and contact networks in shaping infectious disease dynamics. Although these networks share analogous contagion principles, they are often studied in isolation, hindering the incorporation of behavioural, informational, and epidemiological processes into disease models. This review synthesises current research on the interplay between social, mobility and contact networks in health behaviour contagion and infectious disease transmission.

Methods

We searched Web-of-Science and PubMed from January 2000 to June 2025 for research on health behaviour contagion and information dissemination in social networks, pathogen spread through mobility and contact networks, and their joint impacts on epidemic dynamics. This was first done by a preliminary literature screening based on predefined criteria. With potentially relevant publications retained, we performed keyword co-occurrence network analysis to identify the most common themes in studies. The results guide us to narrow down the reviewing scope to the social, mobility and contact network impacts on informational, behavioural, and epidemiological dynamics. We then further identified and reviewed the literature on these multidimensional network influences.

Results

Our review finds that each network type plays a distinct yet interconnected role in shaping behaviours and disease dynamics. Social networks, comprising both online and offline interpersonal relationships, facilitate the dissemination of health information and influence behavioural responses to public health interventions. Concurrently, mobility and contact networks govern the spatiotemporal pathways of pathogen transmission, as demonstrated in recent pandemics. While traditional population-level models often overlook individual discrepancies and social network effects, significant efforts have been made through developing individual-level simulation-based models that integrate behavioural dynamics. With emerging new data sources and advanced computational techniques, two promising approaches—multiplex network analysis and generative agent-based modelling—offer frameworks for integrating the complex interdependencies among social, mobility and contact networks into epidemic dynamics estimation.

Conclusions

This review highlights the theoretical and methodological advances in network-based infectious disease modelling and identifies critical knowledge and research gaps. Future research should prioritise integrating multi-source behavioural and spatial data, unifying modelling strategies, and developing scalable approaches for incorporating multilayer network data. The integrated approach will strengthen public health strategies, enabling equitable and effective interventions against emerging infections.

Individuals’ Self-Disclosure on Social Media Reveals Sentiment and Mental Well-Being Benefits of Vaccination

Wed, 01 Oct 2025 00:00:00 +0000

This project was collaborated with Jianghao Wang, Yichun Fan, and Siqi Zheng

Abstract Vaccination prevents not only the transmission of pathogens but also the adverse psychological effects of epidemics. Prior knowledge of the mental well-being benefits of vaccination relies on cross-sectional surveys, which are limited in scale, subject to self-report bias, and could be confounded by hidden differences across individuals. Here, we collected 154,348 Americans who self-declared receiving the COVID-19 vaccine on Twitter (now called X) and implemented a staggered quasi-experiment design to analyze their 160 million tweet activities, teasing out the causal effect of vaccination on changes in expressed sentiment and mental well-being. We found that being fully vaccinated is associated with an increase in sentiments of 0.055 (95% CI 0.037 to 0.072) standard deviations at the individual level. The effects are stronger among males, adolescents, Hispanic and Asian Americans, and individuals who have experienced severe local epidemics. Investigation of underlying mechanisms suggests that fully vaccinated individuals experience fewer negative emotions, engage in more leisure activities, and are less likely to grapple with conflicting thoughts. Additionally, a more relaxed social environment from lifted lockdowns and increasing herd immunity positively correlates with sentiment improvement. Our study provides a new approach to assessing the impact of medical treatment on mental health by mining large-scale self-reported social media data.

What Digital Mobile Health Systems Add to Infectious Disease Control? A Modelling Study

Mon, 01 Jul 2024 00:00:00 +0000

This project is led by Zhifeng Cheng, Chu He, and Shengjie Lai. We greatly appreciate the Jiangsu Centre for Disease Control and Prevention (CDC) for developing and operating the mHealth system, and for sharing the data.

Abstract Digital health tools, including mobile device location-based health (mHealth) applications, have been adopted to augment conventional contact tracing during recent epidemics and pandemics, yet quantitative evidence about their added epidemiological value remains limited. Using aggregated, privacy-preserving mHealth data from six Chinese cities (July 2021 - May 2022), we modelled transmission dynamics of emerging respiratory infections to quantify the heterogeneous effects of mHealth-triggered movement restrictions and risk-based quarantines across viral variants. We found that the mHealth-assisted interventions, together with population-wide lockdowns, could rapidly reduce the effective infection rate and drive daily incidence toward zero within two weeks, which was rarely achieved through traditional contact tracing alone. Scenario simulations further showed that in the absence of mHealth interventions, infections consistently propagated to connected cities across a wide range of epidemiological assumptions, contradicting the reality. These results suggest that early integration of geolocated mHealth tools with traditional epidemiological investigations enhances outbreak containment at both local and regional scales, providing quantitative, spatially explicit evidence to inform scalable mHealth integration strategies for future infectious threats with pandemic potential.

The Heterogeneous Impacts COVID-19 Pandemic on Global Scientific Collaboration

Sun, 23 Jul 2023 00:00:00 +0000

This project was collaborated with Jianghao Wang and Xiaofan Liang

Abstract Scientific collaboration catalyzes knowledge diffusion and breeds innovation. However, the COVID-19 pandemic has disrupted the rhythm of research cooperation. To track the dynamics and key drivers of scientific collaboration after the COVID-19 outbreak, we analyzed a scientific collaboration network from 25.2 million publications worldwide from 2010 to 2022. We found that the volume of non-COVID-related international collaboration declined 22.2% by late 2022 compared to the pre-pandemic growth trend, with greater impacts on social science, large teams, and top schools. Comparatively, COVID-related international collaboration increased by 179.7%. The scholar-level analysis revealed that researchers developed fewer new collaborators, shifted their research interests to COVID-related topics, and junior academics were more affected. Despite remarkable changes, the impact of the pandemic on international collaborations is less conspicuous than on domestic collaborations. Our findings suggest that funders and institutes should balance rapid funding for disciplines and scientists being most disrupted to recover from the pandemic.

When geostatistics meets machine learning: how do they compete and complement in dynamic population mapping?

Fri, 05 Aug 2022 00:00:00 +0000

This project was led by Zhifeng Cheng and Jianghao Wang.

Abstract Fine-grained dynamic population data are instrumental in climate change response, resource allocation, and epidemic control. With the increasing availability of high-frequency human digital footprint data, the past decades have witnessed numerous efforts in mapping populations at fine spatiotemporal scales. Two methodological families dominate this space: spatial statistical models grounded in geostatistics, which explicitly encode spatiotemporal dependency, and machine learning techniques, which flexibly leverage rich auxiliary covariates. However, the field still lacks a unified standard in modelling strategy and auxiliary data selection, particularly a systematic comparison between newly developed machine learning techniques and traditional spatial statistical methods under different covariate provisions.

This project addresses both gaps. In a case study of hourly population mapping at 100-m resolution in Beijing, we benchmarked two spatial statistical models, the Bayesian space-time model and geographically and temporally weighted regression, against two machine learning techniques, random forest and eXtreme gradient boosting. The Bayesian space-time model in conjunction with urban function data achieved the best performance, underscoring the value of spatiotemporal dependency information and urban function covariates in fine-scale population mapping. Building on these insights, we constructed a hybrid downscaling model that integrates random forest with area-to-point kriging, and applied it to time-series mobile phone positioning data to produce monthly 1-km gridded population distributions across China in 2015. The estimated products capture inner-annual population variations, especially during periods of large-scale population movement such as festivals, holidays, and short-term labour flows, and exhibit higher consistency with census data than popular global population products. Together, the hourly Beijing maps and the monthly nationwide China dataset are released to the research community as a credible foundation for population-dependent decisions.