Labor mobility buffers cascading supply-chain losses from occupational heat stress Authors: Qianzi Wang, Qi Zhou, Shen Qu Published: 2026 ABSTRACT Escalating global temperatures threaten economic stability by worsening occupational heat stress and reducing workforce productivity. Despite advancements in macroeconomic modeling, current risk assessments rely on coarse annual aggregations and ignore internal labor mobility, thereby masking highly unequal sub-national vulnerabilities and underestimating how labor mobility buffers cascading supply-chain losses. Here we present a high-resolution, agent-based dynamic supply chain network model that integrates empirical daily mobility data across 313 Chinese cities to quantify the spatiotemporal cascading economic impacts of occupational heat exposure. We show that annual heat stress costs China 2933.5 billion CNY (2.6% of GDP), with systemic propagation through supply chains driving 59% of these losses. Crucially, labor mobility redistributes risk: net labor inflows into industrialized, high-heat southeastern regions provide a factor-compensation effect that buffers cascading losses by offsetting direct local productivity shocks, saving a net 7.2 billion CNY directly and 24.6 billion CNY indirectly nationwide. Under a 2030 warming scenario (SSP3-7.0), total losses expand 1.6-fold to 4672.9 billion CNY, though integrated multi-level adaptations—combining industrial restructuring with work-hour shifting—can mitigate these future losses by 30%. These findings reveal that demographic mobility dictates the economic geometry of climate vulnerability, highlighting that resilient climate adaptation requires synchronized network-level interventions rather than isolated local policies. Introduction Rising temperatures and excessive heat exposure caused by climate change pose major threats to multiple dimensions of human life, including socioeconomic stability, environmental integrity, and public health [1]. According to the Sixth Assessment Report released by the Intergovernmental Panel on Climate Change in 2023, scenarios projecting 1.5, 2, and 4 °C rises in average global temperatures could increase the frequency of extreme heat waves by 4.1, 5.6, and 9.4 times, respectively [2]. This escalation in heat waves significantly increases the risk of heat stress, particularly occupational heat stress from exposure to elevated workplace temperatures. Previous studies have demonstrated that heat stress consistently pushes labor productivity and human health to precarious thresholds, adversely affecting workforce efficiency and worker well-being [3]. According to conservative estimates by the International Labour Organization, heat stress will result in a loss of 2.2% of total working hours globally by 2030, amounting to economic losses of $2.4 trillion [4]. To assess these economic impacts, it is essential to consider key factors, such as heat stress indicators, heat exposure–response functions, and industrial economic models. Early studies on heat stress predominantly focused on single temperature indicators, such as maximum temperature [5], mean temperature [6], perceived temperature [7], or the number of heat wave days [8], but overlooked other factors affecting human well-being, such as humidity and radiation. Consequently, current research favors composite indices for characterizing heat stress intensity, with wet-bulb globe temperature (WBGT) as the most widely used metric in occupational safety and health studies [9,10]. Exposure–response functions linking heat stress metrics to direct labor productivity quantify productivity loss rates [11,12]. Commonly used exposure–response functions include those used in the High Occupational Temperature Health and Productivity Suppression (Hothaps) program [13,14] and those developed by the International Organization for Standardization [15]. Localized losses in production capacity may propagate through supply chain networks, causing cascading impacts. Pioneering research in this domain has employed industrial economic models to analyze impacts, drawing on historical meteorological and economic datasets [5]. For example, Szewczyk et al. used a macroeconometric model for Europe to estimate the monetary value of productivity changes attributable to heat stress [16]. Input–output (IO) and computable general equilibrium (CGE) models are frequently used to assess macroeconomic losses. For example, Xia et al. quantified the impact of a 14-day heat wave in Nanjing, China, in 2013 using an IO model [8]. Knittel et al. used a CGE model to examine the consequences of climate-change-induced productivity losses for German trade in imports and exports [3]. Although IO and CGE models are useful for assessing direct and indirect losses, they have limitations. IO models often assume constant technology coefficients and ignore factors such as relative prices, the ability of supply chain actors to adjust production, and technology improvements. Therefore, they may underestimate economic resilience and overestimate economic consequences [17]. CGE models are suitable for long-term scenarios and seek optimal solutions through price adjustments; however, they are unsuitable for assessments requiring high temporal precision [18]. Recently, adaptive agent-based supply chain network models have emerged as more effective tools, offering high-resolution spatiotemporal simulations that capture dynamic economic outcomes and cascading effects in real-world supply chains [19,20]. Despite these advancements, assessments of occupational heat stress under climate change have several constraints. First, economic models typically operate at annual and national scales, resulting in a significant loss of granularity in the daily climate data they use. Second, many analyses focus on global economic impacts without examining unequal distribution within individual countries, such as China. Finally, existing studies in this area have overlooked the role of labor mobility. Advances in transportation have significantly reduced the time and cost of mobility, fostering highly mobile societies. This increased mobility indirectly influences heat stress risks by altering occupational structures and increasing pressure in heat-stressed work environments. Thus, it is crucial to consider labor mobility when studying occupational heat stress, as it is a significant factor in the dynamics of heat exposure. To address these shortcomings, we propose a high-resolution dynamic supply chain network model based on adaptive agents that incorporates daily labor mobility into the heat exposure assessment framework. This model evaluates both the direct and indirect (cascading) economic impacts of daily occupational heat stress, with a particular focus on how labor mobility alters heat exposure risks. Labor mobility data were sourced from the Baidu Migration Platform, which provides billions of daily location records covering a significant portion of the Chinese population. Daily city-level WBGT was used as a heat stress indicator, with data derived from 0.25° × 0.25° gridded satellite and meteorological data from the ERA5 hourly reanalysis database of the European Centre for Medium Range Weather Forecasts for China over 365 days. These data were coupled with the Hothaps exposure–response function to estimate the local labor productivity loss rate attributable to heat exposure. These local loss rates were then integrated into a dynamic supply chain network model to evaluate the daily cascading economic losses of occupational heat stress throughout the year. We analyzed multiple scenarios involving labor mobility, adaptive strategies, and climate change to evaluate how these factors influence productivity losses and the propagation through the supply chain [21]. We then conducted high-precision spatiotemporal simulations to explore the heterogeneity of losses across regions and sectors in China, providing valuable insights for policymakers and stakeholders. Materials and methods Research framework This study employed an interdisciplinary approach to evaluate economic losses associated with occupational heat exposure while accounting for labor mobility. First, daily indoor and outdoor wet-bulb globe temperature (WBGT) values were derived from 0.25° × 0.25° gridded climate data and aggregated to regional population-weighted estimates using gridded population data. We then incorporated labor mobility into the exposure–loss framework. Since direct labor mobility data were unavailable, we used population mobility data as a proxy. Next, we estimated labor productivity losses using the WBGT and heat exposure–response functions and categorized them by labor intensity: high-intensity labor was treated as outdoor work, whereas moderate- and low-intensity labor were treated as indoor work. For moderate- and low-intensity labor, we adjusted productivity losses by accounting for air-conditioning coverage. Finally, we evaluated the industrial economic impact. We fed city-level labor productivity losses into the self-developed dynamic supply chain model to simulate daily economic losses and used adaptation and climate change scenarios for the scenario analysis. Data We obtained hourly climate data, including hourly maximum 2 m temperature after post-processing, 2 m temperature, and 2 m dewpoint temperature, from the ERA5 hourly reanalysis database at 0.25° × 0.25° resolution (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=form). Gridded population data for 2020 at the same resolution were obtained from the National Aeronautics and Space Administration database (https://sedac.ciesin.columbia.edu/data/set/gpw-v4-population-count-adjusted-to-2015-unwpp-country-totals-rev11/data-download), and city-level population data for China in 2021 were obtained from the LandScan database (https://landscan.ornl.gov/metadata). City-level daily mobility data were sourced from the Baidu Migration Platform (https://qianxi.baidu.com). Economic impact simulations used the 2017 China multi-regional IO table, the latest available version from the Carbon Emission Accounts and Datasets database (https://www.ceads.net.cn/). For the climate change scenario, we obtained 2030 climate projections under Shared Socioeconomic Pathway 3-7.0 (SSP3-7.0) from the Climate Change Heat Impact and Prevention (CHIP) website (http://www.climatechip.org/). These projections are based on multiyear climatological averages from the Coupled Model Intercomparison Project Phase 6 framework, reducing the influence of high-frequency interannual climate variability. Regional daily population-weighted WBGT We used WBGT to quantify environmental heat stress, calculating separately for indoor and outdoor work environments. Individual-level heterogeneity in heat exposure and sensitivity was not considered. Indoor WBGT was calculated following the method proposed by Epstein and Moran [22] and Stull [23], which assumes a linear relationship between WBGT and wet-bulb temperature (equation (1)). Outdoor WBGT was estimated using a simplified WBGT formulation [24] that accounts for temperature and humidity but excludes wind and radiation effect (equation (2)). Because an 8-h workday is common in China, we used the mean hourly WBGT over a representative work period to characterize daily occupational heat exposure. This period was defined as 9:00 a.m. to 6:00 p.m. local time, excluding a 1 h lunch break from 12:00 p.m. to 1:00 p.m. We then matched the 0.25° × 0.25° WBGT grids with population grids at the same resolution. Grid-level WBGT values was weighted by population and aggregated to the regional scale (equation (3)):(1)Win,i,d=∑h[0.7Tw(Ri,h,Ti,h)+0.3Ti,h]8(2)Wout,i,d=∑h[0.567Ti,h+0.393ρ(Ri,h,Ti,h)+3.94]8(3)Wr,d=∑i∈rWi,dPi∑i∈rPiwhere d denotes the day of year; h is the working hours on day d over an 8-h period; i indexes the grid cell; and r denotes the region. Win,i,d and Wout,i,d are the indoor and outdoor WBGT (°C) on day d in grid cell i; Ti,h is the 2-m air (dry-bulb) temperature (°C) at hour h in grid cell i; Ri,h is the corresponding relative humidity. Tw(Ri,h,Ti,h) is the wet-bulb temperature (°C) and ρ(Ri,h,Ti,h) is the water-vapour pressure (hPa); both were calculated from Ri,h (via the Magnus–Tetens approximation [25]) and Ti,h, with the detailed expressions provided in Supplementary Text S1. Pi is the population of grid cell i in 2021, adjusted using 2020 data. In equation (3), Wi,d denotes either Win,i,d or Wout,i,d, and Wr,d is the corresponding population-weighted WBGT (°C) of region r on day d. Air-conditioning coverage rate We accounted for air-conditioning coverage when estimating labor productivity in indoor environments, because air conditioning can reduce heat exposure and is an important adaptation measure in high-temperature workplaces. The prevalence of air-conditioning equipment depends on climatic conditions and economic development [26] and was therefore represented by the product of saturation and accessibility (equation (4)). Saturation was calculated from cooling degree days (CDDs), defined as the annual sum of daily temperature exceedances above the cooling reference temperature (equation (5)). In China, this reference temperature is commonly set to 26 °C. Accessibility is a function of income level (equation (6)). Exponential and logistic functions have been used to calculate saturation and accessibility [27], and local parameters are generally used for calibration [28]. We calibrated the parameters using the observed penetration rate of air-conditioning equipment, CDD data, and China's per capita gross domestic product (GDP) from 1978 to 2018, following the methods used by Takakura [29]. We assumed that indoor workers experienced no heat-related productivity loss when air conditioning was available and did not include the economic costs of air-conditioning use.(4)Cr=SrAr(5)Sr=1−α1exp(α2Dr)(6)Ar=11+exp(β1−β2Ir)where Cr is the air-conditioning coverage (penetration rate) of region r, calculated as the product of the saturation Sr and the availability Ar; α1 (0.970), α2 (−0.00505), β1 (3.086), and β2 (0.150) are local parameters governing the probability of air-conditioning devices ownership, taken from Liu et al. [28]; Dr is the number of CDDs in region r, with calculation provided in Supplementary Text S1; and Ir is the regional income, expressed as labor earnings in thousand CNY, in region r. Productivity loss ratios In this study, labor productivity is defined as the proportion of a working hour in which a worker effectively performs their intended tasks; work capacity is 100% when heat exposure does not require rest. The functional form of the relationship between labor productivity and WBGT exposure is based on the Hothaps program, which has been widely used by scholars. We adopted Liu et al.’s parameterization, which was recalibrated using epidemiological data from China [28]. A two-parameter logistic function with intensity-specific parameters was used to evaluate the relationship between labor productivity and WBGT across high, moderate, and low labor intensities (equation (7)) [30]. Following Liu et al.’s classification, we assigned the 42 sectors in China, defined in the CEADs database [31], to high, moderate, and low labor intensities (Supplementary Table S1) and estimated daily labor productivity for the sectors on a regional scale:(7)ηr,s,d=0.1+0.91+(Wr,dγ1)γ2where s represents the sector; ηr,s,d is the labor productivity of workers in sector s in region r on day d; and γ1 and γ2 are the estimated parameters for different labor intensities: 34.64 and 22.72 for the low-intensity sector, 32.93 and 17.81 for the moderate-intensity sector, and 30.94 and 16.64 for the high-intensity sector, respectively. The productivity loss ratio was calculated using the first-order effects method [32], in which labor productivity is adjusted to account for the productivity loss ratio (equation (8)), using the ratio of employee compensation to value added across sectors within each region. High-intensity work refers to outdoor work. Air conditioning was considered for low- and moderate-intensity work, which was assumed to be indoor work (equation (9)).(8)Lout,r,s,d=(1−ηr,s,d)Er,s,dVr,s,d(9)Lin,r,s,d=(1−ηr,s,d)Er,s,dVr,s,d×(1−Cr)where Lout,r,s,d and Lin,r,s,d represent the productivity loss ratios for outdoor and indoor work, respectively, in sector s, region r and day d; Er,s,d is the employee compensation (labor income) and Vr,s,d the value added; and Cr is the air-conditioning coverage defined in equation (4). Simulation of industrial economic impact We established a dynamic agent-based model using adaptive agents and complex networks [33]. This model evaluates the direct and indirect (cascading) risks of daily heat-stress exposure on China's industrial economy. It considers labor productivity loss ratios associated with occupational heat exposure at high temporal and spatial resolutions. Direct losses are productivity losses caused by declines in labor productivity from occupational heat exposure, whereas indirect losses arise from the cascading effects of direct losses that spread through the complex upstream and downstream industrial relationships within the supply chain network. Within the supply chain network, this cascading process generates a multiplier effect through two propagation mechanisms: upstream and downstream effects. The upstream effect is driven by information flows. When firms experience productivity losses due to heat exposure, their demand for intermediate inputs declines, reducing orders transmitted to upstream suppliers and leading to lower upstream production. The downstream effect is driven by material flows. When production declines due to heat exposure, the supply of outputs to downstream sectors decreases. If downstream inventories cannot buffer this reduction, production disruptions occur, generating additional indirect losses. Through these mechanisms, initial direct productivity losses propagate across interconnected sectors, amplifying economic impacts throughout the supply chain network. The propagation pathways of these two effects are provided (Supplementary Fig. S1). The model includes production, consumption, and transportation agents that evolve and interact within a complex network, reflecting daily non-equilibrium changes in the economic system. One advantage of the model is that it accounts for agents' adaptive behavior in response to external shocks. This adaptability is reflected in the model's parameters, which can be modified by adjusting factors such as inventory levels, excess production capacity, trade partners, production technology, and post-disaster reconstruction efforts. Heat exposure is treated as an exogenous shock. After the productivity loss ratios were calculated, they were entered into the steady-state system, which was initialized by the complex supply chain network model. The model was used to conduct daily simulations for 365 days to generate output values to calculate the direct and indirect losses to the industrial chain in China. The theoretical basis and mechanisms of the model are detailed in Supplementary Text S2. Labor mobility adjustment Labor mobility can alter regional labor exposure and the economic losses associated with occupational heat stress. In short-term production, the net outflow of labor reduces labor as a production factor in the region, increasing losses and decreasing productivity. Conversely, the net inflow of labor has the opposite effect. This method applies only to short-term production because the complex network model utilizes a Leontief production function, in which production factors have fixed proportions. In contrast, for long-term production, diminishing marginal returns, technological constraints, reduced complementarity effects, and other production factors influence the impact of labor on productivity. It was assumed that labor mobility and population mobility trends were similar. When considering labor mobility, to exclude non-work-related mobility, such as family visits and tourism, we summarized daily labor mobility data on a monthly scale. To verify the robustness of using monthly labor mobility data, we compared the distribution of losses across regions and sectors under daily and monthly labor mobility (Supplementary Fig. S2). The results show that daily labor mobility may overestimate the proportion of labor and consequently underestimate direct losses. However, the difference in total losses between the daily and monthly scenarios was only 0.3%. In terms of the number of affected areas, the number of regions with increased loss under the two labor-flow modes was the same. At the sectoral level, both mobility aggregation schemes increased direct and indirect losses across most sectors. To further examine the robustness of the labor mobility data to changes in temporal aggregation, we also aggregated labor mobility to seasonal and annual scales. The results showed deviations from the business-as-usual (BAU) scenario of 3.5% and 4.5% on the seasonal and annual scales, respectively. The robustness comparison across daily, monthly, seasonal, and annual scales is provided (Supplementary Fig. S3). As an external check of the mobility measure, we compared annual Baidu net flows with the official outward labor rate and found consistent labor migration patterns (Supplementary Fig. S4). Equations (10), (11), (12) were used to perform dynamic population adjustments. After incorporating labor mobility, the productivity loss ratios were adjusted using the dynamic labor adjustment coefficient:(10)λmob,r,d=Pmob,r,dP2021,r(11)Lout,mob,r,s,d=Lout,r,s,dλmob,r,d(12)Lin,mob,r,s,d=Lin,r,s,dλmob,r,dwhere Pmob,r,d is the remaining (mobility-adjusted) population of region r on day d, with the calculation provided in Supplementary Text S1; P2021,r is the 2021 population of region r; λmob,r,d is the dynamic labor-mobility adjustment coefficient for region r on day d; and Lout,mob,r,s,d and Lin,mob,r,s,d are the mobility-adjusted outdoor and indoor productivity-loss ratios of sector s in region r on day d. Scenario analysis We used scenario analyses to assess the effects of labor mobility, climate change, and adaptation on occupational heat-stress losses (Table 1). The BAU scenario was used to assess the daily economic losses attributable to occupational heat stress in the supply chain from September 13, 2021, to September 12, 2022, considering labor mobility. The counterfactual (CF) scenario excludes labor mobility. In the context of climate change, we used the 2030 climate change scenario (S0) to investigate future occupational heat stress losses, assuming that labor mobility in 2030 would be consistent with the BAU scenario. We also tested three adaptation scenarios: national-level labor restructuring (S1), enterprise-level work-hour adjustment (S2), and a multilevel adaptation (S3) that combined S1 and S2.Table 1Scenarios.Table 1Scenario categoryScenario nameDescriptionBusiness-as-usual scenarioBAUDaily occupational heat exposure from September 13, 2021, to September 12, 2022, including the impact of labor mobility.Counterfactual scenarioCFSame as the BAU scenario, but excluding labor mobility.2030 climate change scenarioS0: 2030 climate changeOccupational heat-stress losses in 313 regions of China in 2030, estimated from WBGT projected using Climate CHIP under SSP3-7.0, assuming unchanged labor mobility patterns.2030 adaptive scenariosS1: National-level labor restructuringBased on the S0 scenario. Reflects national labor-market adjustments, with labor demand shifting from high-intensity industries to lower-labor-intensity industries. Relative to BAU, the labor productivity loss rate is predicted to be 41.7% lower in the primary industry, 20% lower in the secondary industry, and 29.0% higher in the tertiary industry.S2: Enterprise-level work-hour adjustmentBased on the S0 scenario. Represents enterprise-level strategies through work-hour shifts to avoid the hottest part of the day, reducing employee heat exposure. Peak working hours from 11:00 a.m. to 12:00 p.m. and 1:00 p.m. to 2:00 p.m. were shifted to the cooler period of 6:00 p.m. to 8:00 p.m.S3: Multi-level adaptationCombines labor restructuring and work-hour adjustment to evaluate their joint effect on heat-stress lossesWBGT: wet-bulb globe temperature; CHIP: Climate Change Heat Impact and Prevention database; SSP3-7.0: Shared Socioeconomic Pathway 3-7.0. Table 1Scenarios.Table 1Scenario categoryScenario nameDescriptionBusiness-as-usual scenarioBAUDaily occupational heat exposure from September 13, 2021, to September 12, 2022, including the impact of labor mobility.Counterfactual scenarioCFSame as the BAU scenario, but excluding labor mobility.2030 climate change scenarioS0: 2030 climate changeOccupational heat-stress losses in 313 regions of China in 2030, estimated from WBGT projected using Climate CHIP under SSP3-7.0, assuming unchanged labor mobility patterns.2030 adaptive scenariosS1: National-level labor restructuringBased on the S0 scenario. Reflects national labor-market adjustments, with labor demand shifting from high-intensity industries to lower-labor-intensity industries. Relative to BAU, the labor productivity loss rate is predicted to be 41.7% lower in the primary industry, 20% lower in the secondary industry, and 29.0% higher in the tertiary industry.S2: Enterprise-level work-hour adjustmentBased on the S0 scenario. Represents enterprise-level strategies through work-hour shifts to avoid the hottest part of the day, reducing employee heat exposure. Peak working hours from 11:00 a.m. to 12:00 p.m. and 1:00 p.m. to 2:00 p.m. were shifted to the cooler period of 6:00 p.m. to 8:00 p.m.S3: Multi-level adaptationCombines labor restructuring and work-hour adjustment to evaluate their joint effect on heat-stress lossesWBGT: wet-bulb globe temperature; CHIP: Climate Change Heat Impact and Prevention database; SSP3-7.0: Shared Socioeconomic Pathway 3-7.0. Scenarios. WBGT: wet-bulb globe temperature; CHIP: Climate Change Heat Impact and Prevention database; SSP3-7.0: Shared Socioeconomic Pathway 3-7.0. BAU and counterfactual scenarios In the BAU scenario, we estimated daily direct and indirect losses caused by occupational heat stress from September 13, 2021, to September 12, 2022. We considered monthly labor mobility between regions to exclude the impact of short-term non-labor mobility, such as cross-regional tourism and family visits. Population mobility data were sourced from the Baidu Migration Platform, which provides billions of daily location records covering a significant portion of the Chinese population, enabling robust tracking of population mobility. The Baidu location data are user-authorized and represent real-time population movements across cities. We utilized public city-level inflow and outflow migration data from Baidu Migration Big Data to estimate monthly intercity inflows and outflows. We established the CF scenario, which is based on the BAU scenario but excludes labor mobility, and then compared the results of the two scenarios to assess the impact of labor mobility. Climate change and adaptive scenarios We established a 2030 climate change scenario (S0) to investigate the impact of future temperature changes on heat exposure. In this scenario, we calculated WBGT for 313 regions in China using temperature projections from Climate CHIP under the SSP3-7.0 pathway, and then estimated the associated heat-exposure losses from occupational heat stress. Adaptive strategies, such as self-adaptation and planned adaptation, are required to mitigate occupational heat stress losses in the context of climate change. Self-adaptation is an unconscious response to high temperatures, such as reducing work intensity, resting more, and drinking more water. Planned adaptation refers to policy decisions informed by an understanding of the adverse effects of heat, such as choosing a lower-intensity job, changing working hours, and similar strategies [34]. Self-adaptation is difficult to quantify in our model; therefore, we focused on planned adaptation. The S1 scenario is based on national-level labor restructuring. The Research Group on Population Structure Changes and Employment Situation at the Development Research Center of the State Council predicted China's labor supply and demand trends over the next decade [35]. Labor demand is predicted to decline in primary and secondary industries, which are sectors with moderate and high labor intensity. In contrast, it is expected to increase in tertiary industries with lower labor intensity. Therefore, we adjusted the labor productivity loss rate for the three major industries in response to changes in labor demand. Companies also adopt adaptive measures to mitigate the impact of heat exposure, such as adjusting working hours to avoid high-temperature periods [36]. To reflect this enterprise-level adaptation, we created an enterprise-level work-hour adjustment scenario (S2). Peak working hours coinciding with the hottest period (11:00 a.m. to 12:00 p.m. and 1:00 p.m. to 2:00 p.m.) were shifted to the cooler hours of 6:00 p.m. to 8:00 p.m. The multilevel adaptation scenario (S3) combines labor-intensity restructuring in S1 and work-hour shifting in S2. The settings for the S1–S3 adaptation scenarios are described in Text S3 in the Supplementary Materials. In S1, we assumed no long-term technological change and therefore did not account for productivity changes arising from technological progress. Results Economic loss assessment of occupational heat exposure We considered several key scenarios to evaluate economic impacts under different conditions. Direct losses and net population mobility changed across regions with different levels of economic development under the BAU scenario compared with the CF scenario (Fig. 1a). A geographic map further presents direct loss changes and net labor mobility in China (Supplementary Fig. S5). By the end of the study period, most regions with net labor inflows (red bubbles) were economically developed, including the southeastern coast of China, the Yangtze River Delta, and the Pearl River Delta. In contrast, regions with net labor outflow (blue bubbles) were mostly underdeveloped areas, such as northern, northeastern, and northwestern China, where temperatures are relatively low. Nearly all regions with a net labor inflow experienced reduced direct loss, corresponding to the red area at the bottom right (Fig. 1a). This pattern suggests that labor inflows increase short-term labor input and production capacity, which may partially offset heat-induced productivity losses.Fig. 1Geographic distribution and breakdown of economic losses and labor mobility. a, Scatterplot of changes in direct losses and net labor mobility. The size of the circles represents the regional gross domestic product (GDP). The blue shading indicates regions with a net outflow of labor and increased direct losses, whereas the red shading indicates regions with a net inflow of labor and decreased direct losses. b, Waterfall chart of the distribution of direct, indirect, and total losses across scenarios. Grey bars depict the total losses across scenarios; light green bars represent between-scenario differences in indirect losses, and dark green bars represent between-scenario differences in direct losses. CF loss: counterfactual loss; Loss (+) indicates an increase in losses; Loss (−) indicates a decrease in losses; BAU: business-as-usual. Geographic distribution and breakdown of economic losses and labor mobility. a, Scatterplot of changes in direct losses and net labor mobility. The size of the circles represents the regional gross domestic product (GDP). The blue shading indicates regions with a net outflow of labor and increased direct losses, whereas the red shading indicates regions with a net inflow of labor and decreased direct losses. b, Waterfall chart of the distribution of direct, indirect, and total losses across scenarios. Grey bars depict the total losses across scenarios; light green bars represent between-scenario differences in indirect losses, and dark green bars represent between-scenario differences in direct losses. CF loss: counterfactual loss; Loss (+) indicates an increase in losses; Loss (−) indicates a decrease in losses; BAU: business-as-usual. This pattern reflects two opposing mechanisms through which labor mobility influences economic outcomes. On the one hand, labor inflows increase firms’ short-term labor input, thereby enhancing production capacity and partially offsetting productivity losses. On the other hand, increased heat exposure experienced by workers reduces firm-level productivity. The observed economic effects arise from the interaction of these two countervailing forces (Supplementary Fig. S6). The larger circles in the red area than in the blue area indicate that most regions with net labor inflows (reduced direct losses) are more economically developed (Fig. 1a). This pattern reflects employment-driven migration toward economically developed regions. The net inflow of labor reduced direct losses in the southeast coastal area, thereby alleviating the direct impact of occupational heat stress there (Fig. 1a). However, underdeveloped regions, such as the north, experienced greater economic losses from net labor outflows due to their relatively small economic size, thereby exacerbating direct losses. Therefore, regions with higher temperatures experienced greater compensation effects from labor mobility than those with lower temperatures, mitigating the unequal impact of occupational heat stress across regions. Economic losses differed between the BAU and CF scenarios (Fig. 1b). Under BAU, the nationwide loss was 2933.5 billion Chinese yuan (CNY), approximately 2.6% of China's GDP in 2021. Indirect losses propagated through the supply chain network accounted for 1729.6 billion CNY, constituting 59% of total losses. This finding underscores the significant indirect impact of occupational heat exposure, particularly through supply chain transmission. Total losses in the CF scenario were 2965.4 billion CNY—1.1% higher than losses in the BAU scenario. The CF scenario, which excludes labor mobility, tends to overestimate the total losses caused by occupational heat stress. These results indicate that the majority of labor flows from economically underdeveloped to economically developed areas. Due to differences in economic size, the absolute increase in direct losses from labor outflow is smaller than the absolute decrease in direct losses from labor inflow. Combining the two opposing effects resulted in a net reduction of 7.2 billion CNY in direct losses in China (Fig. 1b). This indicates that labor mobility reduced the nationwide direct loss attributable to heat stress. Indirect losses decreased by 39.6 billion CNY in some regions but increased by 15.0 billion CNY in others. The 7.2 billion CNY change in direct losses led to a more than threefold change in indirect losses (24.6 billion CNY) due to the multiplier effect of direct loss propagation in the supply chain network. This result suggests that changes in direct losses under the mobility scenarios may have been concentrated in regions with greater supply chain centrality. According to this labor mobility pattern, compensation for labor mobility offsets the direct impact of occupational heat stress on economically developed regions in the southeast. However, these effects were exacerbated in less developed regions. Furthermore, the decline in direct losses reduced indirect losses, lowering the total economic and social costs of occupational heat stress in China. Evolution of daily economic losses Economic losses were simulated daily to track the loss trend. Occupational heat exposure substantially affected China's daily economic losses under the BAU scenario (Fig. 2). Significant indirect and direct losses occurred primarily in the summer months, particularly July and August (Fig. 2a). The trends of total, direct, and indirect losses were highly similar to the daily WBGT trend throughout the year. When WBGT was lower in autumn and winter, the relative share of indirect losses increased, while the share of direct losses decreased. The highest total loss occurred on July 15, 2022 (38.7 billion CNY), accounting for 11.7% of the average daily GDP in 2021. In addition, the peaks of direct and indirect losses were consistent, and the indirect loss peak generally lagged the direct loss peak. This lag occurred because agents in the supply chain network were adaptive and alleviated the impact of external shocks by adjusting inventory, production, and supplier capacities. Since commodity transportation takes time, the generation of indirect losses in downstream agents is delayed.Fig. 2Daily economic losses and their sectoral composition. a, Relationship between daily direct and indirect losses and the wet-bulb globe temperature (WBGT). b, Proportional contributions of indirect and direct losses in high-, moderate-, and low-intensity sectors. Daily economic losses and their sectoral composition. a, Relationship between daily direct and indirect losses and the wet-bulb globe temperature (WBGT). b, Proportional contributions of indirect and direct losses in high-, moderate-, and low-intensity sectors. Sectors with different labor-intensity levels contributed differently to direct and indirect losses (Fig. 2b). High-labor-intensity sectors accounted for 97.8% of the total direct loss. The likely reason concerns the outdoor working conditions in these sectors—that is, workers are exposed to high temperatures without air conditioning, resulting in significant heat-related losses. The moderate-intensity labor sector accounted for 59.2% of total indirect losses. This result suggests that 27 of the 42 sectors fell into the moderate-intensity category, increasing the contribution to indirect losses in the supply chain (Supplementary Table S1). In contrast, the high-intensity sector comprised only two sectors, resulting in only 2.6% of total indirect losses. The losses by sector are analyzed in subsequent sections. Economic losses across regions and sectors Total losses evolved over 365 days across different cities and sectors under the BAU scenario (Fig. 3). The cities and sectors are listed in descending order of total losses. The top 20 regions by total losses included three municipalities and eight provincial capitals (Fig. 3a). Shanghai, the city with the highest loss, recorded 68.6 billion CNY in losses, with indirect losses accounting for 84.9%, or 1.6% of its 2021 GDP. The greatest economic losses in these regions occurred during the summer, with smaller losses in autumn and winter, consistent with national trends. The proportion of indirect losses in these cities ranged from 31.4% (Zhanjiang) to 92.2% (Beijing). Notably, 16 of the top 20 cities had higher indirect than direct losses, with 13 surpassing the national average of 59%. Cities with higher total losses tended to have higher indirect losses, likely reflecting stronger industrial connectivity. Moreover, the timing of peak losses, which coincided with seasonal temperature increases, underscores the importance of assessing seasonal impacts and formulating region-specific response strategies to mitigate the effects of heat exposure on employee health during high-risk periods.Fig. 3Temporal evolution of total, direct, and indirect losses across different regions and sectors. a, Daily evolution and distribution of total losses in the top 20 regions. b, Daily evolution and distribution of total losses across the 42 sectors classified by low (L), moderate (M), and high (H) labor intensity. Different shades represent different degrees of total losses. Dark green bar charts represent direct losses, and light green bar charts indicate indirect losses. Temporal evolution of total, direct, and indirect losses across different regions and sectors. a, Daily evolution and distribution of total losses in the top 20 regions. b, Daily evolution and distribution of total losses across the 42 sectors classified by low (L), moderate (M), and high (H) labor intensity. Different shades represent different degrees of total losses. Dark green bar charts represent direct losses, and light green bar charts indicate indirect losses. Total losses evolved daily across the 42 sectors (Fig. 3b). The seasonal trends in sectoral economic losses were consistent with the national and regional patterns. Agriculture and construction, the two sectors with the highest total losses, are high-labor-intensity sectors, accounting for a combined loss of 1224.0 billion CNY, or 41.7% of total sectoral losses. Workers in these sectors primarily engage in outdoor activities and experience significant occupational heat exposure, contributing substantially to national heat exposure losses. Direct losses exceeded indirect losses in both sectors. Conversely, low- and moderate-intensity labor sectors exhibited higher proportions of indirect losses, exceeding the national average of 59.2%. Primary, secondary, and tertiary industries were affected to varying degrees, with loss proportions of 31.3%, 40.3%, and 28.4%, respectively, corresponding to 0.81%, 1.05%, and 0.74% of GDP. Secondary industries incurred the highest total loss, highlighting the pronounced impact of heat stress on these industries. The distribution of heat-exposure losses across sectors suggests that policymakers should prioritize improving labor conditions in high-intensity sectors, such as agriculture and construction, and address the labor environment in secondary industries. The construction sector exhibited negative indirect losses, indicating model-estimated indirect gains from supply chain reallocation effects rather than a direct benefit of heat exposure. Approximately 93% of the 313 regions experienced negative indirect losses in construction. The construction sector has high industrial connectivity in the supply chain network and benefits from a large number of upstream substitutes, resulting in high flexibility and adaptability to shocks. On the other hand, downstream industries, such as real estate and government entities, primarily influence construction through fixed-asset investments, which tend to remain stable due to their connection to the economic cycle. Thus, the construction sector generated additional revenue through the substitution effect, driven by its higher elasticity relative to upstream industries and stable demand from downstream industries. Adaptation and climate change scenarios Using the scenario framework described in Materials and Methods, we evaluated future heat-stress losses under the 2030 climate-change scenario (S0) and assessed the mitigation potential of three adaptation pathways: labor restructuring (S1), work-hour adjustment (S2) and their combination (S3). Total losses for scenarios S0–S3 and differences in sectoral losses between the S1–S3 scenarios and the S0 scenario were compared (Fig. 4a). Under the S0 scenario, direct and indirect losses are projected to reach 1877.1 billion and 2795.8 billion CNY, respectively, by 2030, corresponding to approximately 1.6 times the total losses under the BAU scenario. Indirect losses are 1.5 times the direct losses, indicating that the cascading effects on the economy are significant and should not be overlooked. The total losses in scenarios S1–S3 are 24.8%, 7.8%, and 30.0% lower, respectively, than in the S0 scenario. The sum of the separate reductions under the S1 and S2 scenarios exceeds the reduction in S3 by approximately 10%, suggesting overlapping effects. The reductions in direct losses in S1–S3 resulted in indirect loss reductions of 1.8, 1.7, and 1.7 times, respectively. This finding suggests that adaptive strategies significantly influence upstream and downstream trade relations in the supply chain network, substantially reducing indirect losses.Fig. 4Scenario-based changes in total losses at regional and sectoral scales. a, Changes in total losses between the S0 and the S1–S3 scenarios. b–c, Changes in total losses in 313 regions between the S0 and BAU scenarios (b) and between the S1–S3 and S0 scenarios (c), with the horizontal axis representing the annual average WBGT for cities in 2030. The top 20 regions in the S0–S3 scenarios are labeled in the scatterplots. The grey dashed lines mark the threshold separating the top 20 regions with the largest changes in total losses from the remaining regions. Scenario descriptions: S0 represents the 2030 climate change scenario under SSP3-7.0; S1 represents national-level labor restructuring; S2 represents enterprise-level work-hour adjustment; and S3 combines S1 and S2. Scenario-based changes in total losses at regional and sectoral scales. a, Changes in total losses between the S0 and the S1–S3 scenarios. b–c, Changes in total losses in 313 regions between the S0 and BAU scenarios (b) and between the S1–S3 and S0 scenarios (c), with the horizontal axis representing the annual average WBGT for cities in 2030. The top 20 regions in the S0–S3 scenarios are labeled in the scatterplots. The grey dashed lines mark the threshold separating the top 20 regions with the largest changes in total losses from the remaining regions. Scenario descriptions: S0 represents the 2030 climate change scenario under SSP3-7.0; S1 represents national-level labor restructuring; S2 represents enterprise-level work-hour adjustment; and S3 combines S1 and S2. The agriculture sector experienced the largest reduction in total losses across all adaptation scenarios (349.4 billion, 99.5 billion, and 425.6 billion CNY in the S1–S3 scenarios, respectively). Agriculture and construction, which are high-intensity labor sectors, consistently ranked among the top two sectors with the largest reductions in all adaptation scenarios. The wholesale and retail sector, considered low-labor-intensive, ranked third in terms of the reduction in total losses. This result may reflect the significant indirect losses in the wholesale and retail sector, which ranked first among the 42 sectors. The wholesale and retail industries are closely connected to other sectors, such as manufacturing, agriculture, and transport, making them particularly vulnerable to supply chain disruptions. Thus, it incurred substantial indirect losses. The 2030 annual average WBGT across all regions was positively correlated with changes in total loss under the S0 scenario, indicating that regions with higher 2030 temperatures will experience greater occupational heat stress losses (Fig. 4b). A consistent trend also appeared for the S1–S3 scenarios, suggesting that regions with higher 2030 temperatures will benefit most from adaptive measures (Fig. 4c). Therefore, in the context of climate change, regions with higher temperatures should be prioritized for adaptation measures. The regions marked in the four scenarios are highly sensitive and should be a focus for policymakers. Among the top 20 regions in all four scenarios, 13 regions overlap, making them particularly critical. These regions are mostly regional core cities and economically strong urban centers, including two municipalities directly under the central government (Beijing and Shanghai) and five provincial capitals (Zhengzhou, Chengdu, Wuhan, Changsha, and Nanning). The full set of overlapping regions is distributed across several key economic areas, including the Yangtze River Economic Belt and the Guangdong–Hong Kong–Macao Greater Bay Area. Many are national transportation hubs (Beijing, Shanghai, Guangzhou, Zhengzhou, etc.) or benefit from port and shipping advantages (Shenzhen, Quanzhou, Zhanjiang, etc.). These transport and logistics functions enable them to build strong industrial-chain connections with surrounding cities. Discussion and conclusion This study used an agent-based complex supply chain network model with high-resolution spatiotemporal data to assess daily direct and indirect losses attributable to occupational heat stress in China over 365 days, from September 13, 2021, to September 12, 2022. The model considers the impact of labor mobility, which has not been systematically analyzed in previous studies. Our research advances understanding of the economic consequences of occupational heat stress, offering several novel insights and methodological improvements. First, our study offers a fine-scale assessment of daily heat-stress risks using an innovative methodology. Advancing high-resolution modeling is crucial for revealing the systematic interactions between extreme weather events and socioeconomic outcomes under global warming [37]. We utilized an agent-based complex supply chain network model tailored to China's city scale to quantify the economic impacts of daily heat stress at a higher temporal resolution than traditional models. Unlike IO or CGE models, which often simplify supply chain dynamics, our framework captures the nuances of adaptive behavior in interconnected networks. These adaptive behaviors, such as adjustments to inventory, production, and supplier capacities, enable dynamic analysis of cascading economic repercussions triggered by heat stress. During the one-year study period, the total losses attributable to occupational heat stress ranged from 2933.5 to 2965.4 billion CNY, with indirect losses accounting for 59.0% to 59.2% of the total, highlighting the pivotal role of cascading effects in amplifying economic disruptions. This finding underscores the criticality of indirect losses in building economic resilience and provides policymakers with a fresh perspective, validating the necessity of fine-grained analysis. Our results provide a robust foundation for developing daily-scale risk alert systems and more precise policy interventions. Second, we systematically analyzed the economic impact of labor mobility on occupational heat stress. Labor mobility significantly reduced the direct losses from occupational heat stress, with indirect losses reduced nearly three times as much as direct losses. In addition, labor mobility altered the geographic and sectoral distribution of occupational heat-stress losses, highlighting unequal impacts. Our results show that labor mobility exhibits a clear spatial pattern toward regions with higher economic development and greater heat stress. Although high-temperature environments in destination areas increase the risk of occupational heat stress per unit of labor, the inflow of labor as a core factor of production significantly enhances local production capacity. This factor effectively offsets, and even exceeds, the output losses from heat, resulting in a decrease in direct economic losses across almost all regions with net labor inflows. In contrast, less-developed northern regions face greater losses due to labor outflows. Therefore, the factor–compensation effect reflects a process of risk redistribution and reshapes both the spatial and sectoral structures of vulnerability. However, population mobility may decrease in the future as China continues its urbanization and balanced economic development [38,39]. As climate change intensifies and associated heat-stress risks rise, this compensatory and redistributive effect may weaken, making it essential to develop targeted adaptation strategies to address emerging inequalities. In addition to the spatial and sectoral heterogeneity identified above, the simulation results also revealed clear temporal variations. Third, our model revealed the significant heterogeneity of losses across regions and sectors in China. High-spatiotemporal-resolution simulations indicated an unequal distribution of heat exposure–induced economic losses across 313 regions and 42 sectors in China. Coastal regions in the southeast bore a disproportionate economic burden. Three municipalities and eight provincial capitals among the top 20 cities with the highest economic losses had indirect-to-total loss proportions ranging from 31.4% to 92.2%, and 13 of these cities had higher proportions of indirect losses than the national average (59.2%). The agriculture and construction sectors, which are labor-intensive and heavily dependent on outdoor work, accounted for 41.7% of the total losses. Although moderate-labor-intensity sectors had relatively low direct losses, their substantial sectoral scale and pivotal positions in supply chains led to the largest indirect losses, accounting for 59% of the annual total. In contrast, high-labor-intensity sectors had more direct losses (98%). These findings highlight the sector-specific nature of occupational heat stress impacts and the need for differentiated strategies to mitigate these effects. Fourth, we evaluated the potential of adaptation strategies to reduce heat stress losses. We considered three scenarios: national-level labor restructuring (S1), enterprise-level work-hour adjustments (S2), and a combined multilevel adaptation scenario (S3). In the S1 scenario, labor distribution was adjusted across industries, resulting in lower labor productivity loss rates in the primary and secondary industries. The total loss was 1158.9 billion CNY less than in the S0 scenario. This adjustment demonstrates the pivotal role of industrial structure optimization in climate change adaptation. In the S2 scenario, adjusting the labor hours to avoid high-temperature periods resulted in a 364.5 billion CNY reduction in total losses. This demonstrates the positive impact of flexible measures at the enterprise level. The combined adaptation scenario (S3), which incorporated the S1 and S2 strategies, achieved a total loss reduction lower than the sum of the individual reductions from S1 and S2, suggesting that the two strategies may overlap or interfere. Nonetheless, the reduction in direct losses from adaptive strategies led to a proportionally larger reduction in indirect losses, underscoring the importance of adaptive policies in mitigating the cascading effects of heat stress. The total losses observed in our study were consistent with previous estimates of GDP losses due to heat stress, which ranged from 0.5% to 4.0% globally [29] and from 1.15% to 2.67% for China [28]. Our results confirm that the agriculture and construction sectors bore the brunt of the impact, aligning with observations by De Sario et al. and Khan et al., who also identified these sectors as highly susceptible to heat stress [40,41]. This effect can be attributed to their labor-intensive nature and heavy reliance on outdoor labor. Previous studies have found that population dynamics alter urban residents' exposure to extreme temperatures in the United States [42]. Labor mobility is pivotal in different domains, from public health to environmental sustainability. It has also been increasingly used to assess risk redistribution across diverse societal domains [43]. In China, population movement during the COVID-19 pandemic affected epidemic spread [44], especially during peak mobility periods such as the Chinese New Year travel season [45]. Guo's study in Australia revealed that mobility affected environmental outcomes: international migration tended to increase carbon emissions in host regions, whereas domestic migration offset them through urban efficiency gains [46]. These findings underscore that mobility is a double-edged sword: it can mitigate or exacerbate systemic stresses depending on regional contexts and policy frameworks. Our study adds a novel climate dimension to this discourse by demonstrating that labor mobility can alter the spatial and sectoral distribution of occupational heat stress risks in a warming climate. Moreover, as Gushulak and MacPherson noted in the context of population health, mobility not only changes exposure but also modifies vulnerability profiles depending on access to healthcare, infrastructure, and adaptive capacity [47]. Our findings resonate with this principle: although the labor movement is a macro-level adaptation, it can increase regional inequality. Thus, our work bridges the fields of public health, environmental policy, and climate economics, expanding the analytical scope of mobility research and underscoring its importance in designing equitable climate resilience strategies. Despite these valuable insights, our study has limitations. First, the production function embedded in our model is the Leontief production function. This aligns with our focus on high-resolution short-term and medium-term dynamics for early warning rather than long-term structural forecasting. Future research could incorporate production functions with factor substitution to examine long-term structural and technological adaptation under climate change. As long-term socioeconomic projection methods continue to improve, future studies may also extend the temporal scope of scenario analysis to mid- or late-century horizons, such as 2050 or even 2100, and incorporate multiple SSP pathways to provide a more comprehensive assessment of long-term climate risks. Second, the WBGT used in this study is based on a simplified formulation that excludes wind speed and solar radiation. While this approach provides a consistent framework for large-scale regional comparisons, it may introduce bias in estimating absolute exposure levels. Future research could integrate comprehensive WBGT formulations that explicitly account for radiation and wind components to improve the accuracy of exposure assessment. In addition, our estimate of air-conditioning coverage did not consider associated costs or additional energy consumption. These costs and environmental implications could be analyzed in future studies. Assuming zero indoor productivity loss is an oversimplification, as fully cooling complex indoor workspaces remains impractical. Moreover, although the economic losses were comprehensively quantified in this study, the analysis did not address the health risks faced by migrant workers. Furthermore, in the enterprise-level adaptation scenario, the assumed feasibility of adjusting working hours may be somewhat optimistic, as institutional, social, and sector-specific constraints should also be considered. Finally, conducting a systematic global sensitivity analysis remains a crucial future direction for further enhancing model confidence. We further discuss the assumptions underlying the labor mobility proxy and possible improvements. For the labor mobility proxy, although the dataset does not distinguish between labor-related and non-labor-related movements, it is currently the most comprehensive high-frequency source for intercity mobility in China in terms of coverage and temporal resolution. Moreover, robustness tests comparing daily and monthly results showed a difference of only 0.3%, and the estimated results for seasonal and annual data indicated minimum deviations from baseline of approximately 3.5% and 4.5%, respectively (Supplementary Fig. S3). This indicates that short-term fluctuations do not drive the core conclusions. Because official population statistics serve as the baseline for the study, the analysis focuses on net interregional changes in labor factors rather than on absolute migration volumes. Future research could incorporate more precise mobility data to refine labor mobility measurements. Finally, the assumption of constant labor mobility trends in the 2030 scenario isolates climate effects, while future long-term studies could apply more advanced methods to forecast high-resolution intercity mobility networks. CRediT authorship contribution statement Qianzi Wang: Writing – review & editing, Writing – original draft, Visualization, Validation, Investigation, Formal analysis. Qi Zhou: Writing – review & editing, Validation, Supervision, Resources, Funding acquisition, Data curation, Conceptualization. Shen Qu: Writing – review & editing, Supervision, Software, Resources, Project administration, Methodology, Funding acquisition. Data availability The climate data were sourced from the ERA5 hourly reanalysis database (https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=form). The latest 0.25° × 0.25° grid population data (2020) were downloaded from the NASA database (https://sedac.ciesin.columbia.edu/data/set/gpw-v4-population-count-adjusted-to-2015-unwpp-country-totals-rev11/data-download). The 2021 city-level China population data were downloaded from the Landscan database (https://landscan.ornl.gov/metadata). City-level daily mobility data were sourced from the Baidu Migration Platform (https://qianxi.baidu.com). The data used in the economic impact simulation were the latest China 2017 multi-regional IO table sourced from the Carbon emission accounts and datasets (CEADs) database (https://www.ceads.net.cn/). In addition, the 2030 climate data (SSP3-7.0) used in the climate change scenario were obtained from the Climate Change Heat Impact and Prevention (CHIP) website (http://www.climatechip.org/). Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.