Marshall Burke’s research while affiliated with Stanford University and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (178)


Fine Particulate Matter From 2020 California Wildfires and Mental Health-Related Emergency Department Visits
  • Article

April 2025

·

2 Reads

JAMA Network Open

Youn Soo Jung

·

Mary M Johnson

·

Marshall Burke

·

[...]

·

Kari C Nadeau

Importance A growing body of research suggests that exposure to fine particulate matter (PM 2.5 ; particle size 2.5 microns or smaller) may be associated with mental health outcomes. However, the potential impact of wildfire-specific PM 2.5 exposure on mental health remains underexplored. Objective To investigate whether wildfire-specific PM 2.5 exposure may be associated with emergency department (ED) visits for mental health conditions, including all-cause and for psychoactive substance use, nonmood psychotic disorders, anxiety, depression, and other mood-affective disorders during the extensive 2020 California wildfire season. Design, Setting, and Participants This cross-sectional study used data on ED visits from July to December 2020 obtained from the California Department of Health Care Access and Information (HCAI). Eligible participants were California residents who presented to an ED in California for mental health conditions without COVID-19. The data were analyzed between July 2020 and December 2020. Exposure Wildfire-specific PM 2.5 exposure (with up to 7-day lags) based on participants’ residential zip codes. Main Outcomes and Measures Daily ED visit counts for all-cause and disease-specific mental health conditions (F00-F99) identified using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) codes at zip code tabulation areas. Results Between July and December 2020, there were 86 609 ED visits for mental health conditions (median [IQR] patient age, 38 [27-54] years; 40 272 female [46.5%]; 10 657 Black [12.3%], 30 044 Hispanic [34.7%], 35 145 White [40.6%]). Visits included psychoactive substance use (23 966 [27.6%]), nonmood psychotic disorders (16 714 [19.3%]), anxiety (26 711 [30.8%]), depression (10 422 [12.0%]), and other mood-affective disorders (5338 [6.2%]). During peak wildfire months, the median (IQR) daily concentration of wildfire-specific PM 2.5 increased to 11.9 (3.9-32.5) μg/m ³ . A 10-μg/m ³ increase in wildfire-specific PM 2.5 was associated with higher ED visits for all-cause mental conditions (cumulative relative risk [cRR] over lag 0-7 days, 1.08; 95% CI, 1.03-1.12), depression (cRR over lag 0-7 days, 1.15; 95% CI, 1.02-1.30), other mood-affective disorders (cRR over lag 0-7 days, 1.29; 95% CI, 1.09-1.54), and anxiety (cRR over lag 0-4 days, 1.06; 95% CI, 1.00-1.12). Subgroup analyses suggested that wildfire smoke was associated with disproportionately increased ED visits among female individuals (eg, depression: cRR over lag 0-4 days, 1.17; 95% CI, 1.03-1.32) and young people (other mood-affective disorders: cRR over lag 0-4 days, 1.46; 95% CI, 1.08-1.98). Effect modification by race was found, with non-Hispanic Black individuals having an increased risk of ED visits for other mood-affective disorders (cRR over lag 0-5 days, 2.35; 95% CI, 1.56-3.53) and Hispanic individuals an increased risk for visits for depression (cRR over lag 0-7 days, 1.30; 95% CI, 1.06-1.59). Conclusions and Relevance Wildfire smoke exposure was associated with significantly increased odds of subsequent ED visits for mental health conditions in this cross-sectional study, with varying lag times for different subconditions and demographic groups.


Quantifying the contributions of climate change and adaptation to mortality from unprecedented extreme heat events
  • Preprint
  • File available

February 2025

·

68 Reads

Understanding the mortality effects of the most extreme heat events is central to climate change risk analysis and adaptation decision-making. Accurate representation of these impacts requires accounting for the effects of prolonged sequences of hot days on mortality, the change in that mortality due to anthropogenic forcing, and the potential compensating effects of adaptation to heat. Here, we revisit the August 2003 heat wave in France, a canonical event in a region with rich climate and mortality data, to understand these influences. We find that standard heat mortality exposure-response functions underpredict excess deaths in August 2003 by 60%, but that accounting for the temporally compounding effects of hot days better matches observed mortality. After accounting for compounding effects and applying a machine learning approach to single-event climate attribution, we attribute 6,038 deaths in August 2003 to climate change, ten times higher than previous estimates. Finally, we show that recent adaptation to heat has reduced the projected death tolls of future 2003-like events by more than 80%.

Download



Improved daily PM2.5 estimates in India reveal inequalities in recent enhancement of air quality

January 2025

·

24 Reads

·

1 Citation

Science Advances

Poor ambient air quality poses a substantial global health threat. However, accurate measurement remains challenging, particularly in countries such as India where ground monitors are scarce despite high expected exposure and health burdens. This lack of precise measurements impedes understanding of changes in pollution exposure over time and across populations. Here, we develop open-source daily fine particulate matter (PM 2.5 ) datasets at a 10-kilometer resolution for India from 2005 to 2023 using a two-stage machine learning model validated on held-out monitor data. Analyzing long-term air quality trends, we find that PM 2.5 concentrations increased across most of the country until around 2016 and then declined partly due to favorable meteorology in southern India. Recent reductions in PM 2.5 were substantially larger in wealthier areas, highlighting the urgency of air quality control policies addressing all socioeconomic communities. To advance equitable air quality monitoring, we propose additional monitor locations in India and examine the adaptability of our method to other countries with scarce monitoring data.


Intensifying risk of mass human heat mortality if historical weather patterns recur

January 2025

·

63 Reads

The potential death toll of worst-case extreme heat events is crucial for climate risk analysis and adaptation planning. We estimate this quantity for Europe using machine learning to calculate the intensity of historical heat waves if they occur at present or future global temperatures, combined with empirical exposure-response functions to quantify the resulting mortality. Each event is projected to generate tens of thousands of excess deaths. For example, if July 1994 or August 2003 meteorological conditions recur at the current global temperature anomaly of 1.5 °C, we project 14,000 or 17,300 excess deaths across Europe in a single week, respectively. At 3 °C, mortality rises to 26,800 or 31,500 per week. These death rates are comparable to peak COVID-19 mortality in Europe and are not substantially reduced by ongoing climate adaptation. Our results suggest that avoiding mass heat mortality in Europe will require significant and novel adaptation to heat.


Fig 2. Impact of Rx fire treatments on burn severity and smoke emissions. (a) All sample estimates
Fig 3. Comparative efficacy of wildfire management strategies. (a) Estimates of burn severity
Fig 4. Net effects and projections of Rx fire treatments on smoke emissions in California.
Fig. S1. Log distribution of acres treated inside the WUI and outside. Mean acreage sizes are reported as
Efficacy of Recent Prescribed Burning and Land Management on Wildfire Burn Severity and Smoke Emissions in the Western United States

December 2024

·

137 Reads

Prescribed fire is increasingly proposed as a policy strategy to reduce wildfire risks, but evidence of its effectiveness in lowering fire severity and smoke emissions remains limited in the western US. We empirically demonstrate that areas treated with prescribed fire and subsequently burned during California’s extreme 2020 wildfire season showed a -14% net reduction in smoke emissions, though these treatments were less effective near populated areas. Our findings suggest that expanding prescribed fire use can meaningfully reduce smoke emissions, even when factoring in smoke from the prescribed fires themselves. The proposed policy of treating one million acres annually in California could reduce overall smoke emissions by 655,000 metric tons over the next five years—equivalent to 52% of the emissions from 2020 wildfires. Our results also suggest that broader application of prescribed fires can provide benefits in mitigating severe wildfire impacts and improve air quality in fire-prone regions worldwide.



Data Summary for countries in our sample. Top panels show the spatial distributions of country level averages and bottom panels show the sample village distribution of (a) average percentage of time experiencing thermal inversions (b) average PM2.5 concentrations and (c) stunting rates for children under 5. Countries not in our sample are colored gray.
Effect of thermal inversion induced increase in PM2.5 on child stunting. (a) the ‘first stage result’ estimated using equation (1) to assess how the total hours of thermal inversions during a pregnancy changes pregnancy average PM2.5. Estimates are shown separately by region and scaled per 100 inversion hours (median sample exposure = 116 inversion hours). Circles indicate point estimates and whiskers indicate 95% confidence intervals. Baseline PM2.5 levels are 40.2 µg m⁻³ for the pooled sample, 34.7 µg m⁻³ for Africa, and 51.3 µg m⁻³ for South Asia. (b) the ‘reduced form’ result from estimating equation (2) shows the estimated percentage change in child stunting from an additional 100 h of thermal inversions during pregnancy. (c) the ‘instrumental variable’ estimates from equation (3) showing the estimated percent change in child stunting per 1-unit increase in PM2.5 during pregnancy induced by additional thermal inversions. Additional details of these results are shown in tables S3–S5.
Effect of thermal inversions by baseline PM2.5. We find that thermal inversion impacts scale with baseline PM2.5. Results show the estimated change in pregnancy average PM2.5 (a) and child stunting rate (b) per additional 100 inversion hours separately by baseline PM2.5 levels in the child’s location of residence. Estimates correspond to equations (8) and (9), respectively (see supplementary materials for details on how these effects are calculated from the regression coefficients). Circles indicate point estimates and whiskers indicate 95% confidence intervals.
Effect of thermal inversions during pregnancy on stunting by child and household characteristics. Results correspond to separate estimates of equation (9) for each category and show how estimated child stunting responses from an additional 100 h of thermal inversions during pregnancy vary by subgroup (see supplemental materials for details on how estimates are calculated from regression coefficients). Colored circles indicate point estimates and whiskers indicate 95% confidence intervals. Black lines indicate statistical significance associated with pairwise comparisons between group-specific estimates ( ∗∗∗: 0.01, ∗∗: 0.05, ∗: 0.1). Absence of line between pair indicates difference between estimates is not statistically different from zero.
Identifying child growth effects of elevated pollution levels during pregnancy

December 2024

·

13 Reads

·

2 Citations

Poor air quality is known to be one of the leading contributors to poor child health globally, and a wealth of evidence has linked pollution exposure during pregnancy to adverse birth and early-life outcomes. While there is some evidence pollution exposure during pregnancy is associated with slowed child growth, this evidence is largely limited to empirical settings in which it is difficult to disentangle the role of pollution from other co-varying factors. Here we rely on quasi-random variation in pollution conditions induced by thermal inversions to estimate the impact of pollution spikes during pregnancy on childhood stunting. We find that thermal inversions during pregnancy worsen air quality and increase the likelihood of childhood stunting, but only in places with poor baseline air quality and particularly for younger children and in lower-wealth communities. Our estimates imply that a 1 µg m⁻³ increase in average PM2.5 concentration during pregnancy increases the probability of stunting by an average of 4.1 percentage points (95% CI: 0.2–8.0). This translates to an 11.2% increase (95% CI: 0.6%–21.9%) in stunting risk from the sample baseline of 37% children stunted. Our results suggest that policies that limit baseline daily PM2.5 levels, particularly during seasons when thermal inversions are more frequent, have potential to generate meaningful improvements in long-run child outcomes.


Figure 3: National and regional patterns in total and non-smoke PM 2.5 show the growing influence of smoke, and continual declines in non-smoke PM 2.5 . Black lines show the national or regional annual average observed PM 2.5 , blue lines show the non-smoke PM 2.5 , and grey shaded areas indicate the portion of PM 2.5 due to smoke. Regional and national annual averages are computed from station annual averages and include only station-years with at least 50 observations. Regions are U.S. climate regions from the EPA (41).
Figure 4: Smoke is increasingly the cause of air quality standard exceedences, for both annual and 24-hour values. a) A station is classified as under threshold if its trailing 3-year average value is below the updated annual average standard (9 í µí¼‡g/m 3 , left panel) or 24-hour standard (35 í µí¼‡g/m 3 , right panel). A station is classified as over threshold without smoke if the 3-year average value of non-smoke PM 2.5 is also over the standard, and as over threshold due to smoke if the value for total PM 2.5 is over the standard but the value for non-smoke is under the standard. b) Air quality monitoring stations are colored by their classification over the last 5 years (2019-2023), with red, blue, and purple indicating that at any time in the last 5 years the station was over threshold due to smoke for annual average, 24-hour daily extremes, or both, respectively.
Figure S3: Change in model performance from adding smoke PM 2.5 interpolations Change in station model performance (measured as change in location-specific í µí± 2 ) from model with AOD predictions and no smoke PM 2.5 interpolations (most closely resembling the features used in Childs et al. (5)) to new preferred model which uses smoke PM 2.5 interpolations and no AOD predictions.
Figure S6: Changes in number of stations and observations per year While the number of stations reporting data each year remains roughly constant, over the time period (2006-2023) an increasing number of stations are reporting over 200 and over 100 observations per year, consistent with a transition from every sixth day (∼60 observations per year) and every third day (∼120 observations per year) reporting to daily reporting.
Figure S8: Since 2021, the length of the "smoke season" has more than doubled compared to 2006-2019 average. Lines show the daily population-average smoke PM 2.5 exposure (vertical axis) throughout the year (horizontal axis), for 2021-2023 (colored lines) and the 2006-2019 average (black line). The 2006-2019 average is calculated as the day-of-year average over the relevant years. We define the "smoke season" as the period of the year stretching from the first to the last time the daily population-average smoke PM 2.5 exposure exceeds 1 í µí¼‡g/m 3 (horizontal dashed line). The duration of the smoke season is shown in horizontal lines, labeled with the year and the length of the smoke season. Points show the 10 worst exposure days in the sample, as in Fig 2. Vertical axis is pseudo-log scaled for visibility.
Growing wildfire-derived PM2.5 across the contiguous U.S. and implications for air quality regulation

December 2024

·

54 Reads

Growing wildfire activity across North America produces significant smoke, undermining efforts to regulate surface air quality and protect public health. Using surface measurements, satellites, and machine learning, we provide granular, daily estimates of smoke PM2.5 concentrations in the contiguous U.S. from 2006 to 2023, and use them to assess the implications of smoke for surface air pollution and its regulation. From 2020 to 2023, population average smoke PM2.5 concentrations were 2.6–6.7 times higher than the 2006–2019 average, with exposure periods twice as long. The 5 worst exposure days in our sample period all occurred in 2023, a year with limited exposure in the western U.S. Wildfire smoke has recently driven 34% of monitoring stations above updated air quality standards, and is making the use of extreme event exemptions challenging. Without wildfire smoke, PM2.5 levels would have continued improving across the US.


Citations (59)


... 8 Further, smoke from wildfires that burned structures appears enriched in heavy metals compared with wildfires that burned vegetation alone. 9 While some of these substances cannot travel long distances, others can as part of smoke. Smoke drives most of the health burden of wildfires due to broad population exposure. ...

Reference:

Beneath the smoke: Understanding the public health impacts of the Los Angeles urban wildfires
The Influence of Wildfire Smoke on Ambient PM2.5 Chemical Species Concentrations in the Contiguous US
  • Citing Article
  • February 2025

Environmental Science and Technology

... Ahmadian et al. explored the complex correlation between outdoor air pollutants and meteorological conditions and further analyzed the potential health risks of this relationship [6]. Kawano et al. proposed a two-stage machine learning model, finding reductions in PM2.5 pollutants were more significant in affluent areas [7]. Other scholars have proposed new frameworks for air quality estimation, such as Zhou's enhanced neural network model incorporating a novel nonlinear autoregressive neural network exogenous input model [8]. ...

Improved daily PM2.5 estimates in India reveal inequalities in recent enhancement of air quality
  • Citing Article
  • January 2025

Science Advances

... Our results may therefore underestimate population exposures. Different methods of estimating smoke PM 2.5 also produce different results (Qiu et al. 2024), suggesting that additional research should examine how these differences impact estimates of cooccurring events and interact with different threshold definitions. Our study suggests several opportunities for future work. ...

Evaluating Chemical Transport and Machine Learning Models for Wildfire Smoke PM 2.5 : Implications for Assessment of Health Impacts
  • Citing Article
  • December 2024

Environmental Science and Technology

... Previous research on climate change and violence has primarily examined violence as an outcome of climate hazards (Burke et al. 2024;von Uexkull and Buhaug 2021;Stechemesser et al. 2021). 1 Within this approach, scholars investigated if, when, and how climate-related hazards shape conflict patterns (Suleymanov 2024;Michelini et al. 2023;Maconga 2023;Hastings and Ubilava 2023;Ide 2023;Cappelli et al. 2023;Mack et al. 2022;Schleussner et al. 2016). Based on this research, there is now scholarly agreement-supported by expert elicitation (Mach et al. 2019), IPCC assessments (IPCC 2023), and broader research syntheses (Beaumont and Coning 2022)-that the climateconflict relationship is highly contextual, socially mediated, and complex. ...

New evidence on the economics of climate and conflict
  • Citing Chapter
  • January 2024

... Both acute and chronic exposure to ambient and household air pollution affects individuals across their life cycle, from prenatal development to old age. For instance, exposure during pregnancy can increase risks of fetal loss, premature birth, and low birthweight [70], while long-term effects include stunting and respiratory issues in childhood [71,72]. In adulthood, air pollution contributes to diseases such as cardiopulmonary conditions, type 2 diabetes, and mental health issues [73][74][75]. ...

Identifying child growth effects of elevated pollution levels during pregnancy

... In the past few decades, as the spatial resolution and spectral resolution of satellite remote sensing images [17,18] have continued to improve [19,20], more and more information [21] can be obtained from remote sensing images [19,22], which can support more sophisticated mapping work [23][24][25][26] and change monitoring [27,28]. The use of ultra-highresolution remote sensing images to study and obtain various urban information has also achieved good results, such as its use in drawing urban land use maps [29][30][31][32], and studying more specific urban road maps [33][34][35][36][37], urban population maps [38], coastal maps [39], urban functional area maps [40,41], urban informal residential area maps [42][43][44], etc. Ultrahigh-resolution remote sensing images provide an important basis for the identification of urban open spaces. ...

Towards transferable building damage assessment via unsupervised single-temporal change adaptation
  • Citing Article
  • December 2024

Remote Sensing of Environment

... First, researchers can subset their data by time period or region to test whether the climate sensitivity varies through time or across space (Hsiang 2016;Kalkuhl & Wenz 2020;Schlenker & Roberts 2009). If sensitivity does not vary, researchers can cautiously conclude that acclimation or adaptation may not be playing a strong role (Burke et al. 2024;Hsiang 2016;Schlenker & Roberts 2009), at least within the observed data and scale. Second, researchers can include a time-period-by-climate interaction in their model to test if climate sensitivity changes over time (Dudney et al. 2021). ...

Are We Adapting to Climate Change?
  • Citing Article
  • January 2024

SSRN Electronic Journal

... In an extensive analysis of 73 million individuals based on Medicare data, decreasing PM2.5 levels from 12 µg per cubic meter to 8 µg per cubic meter had a more significant reduction in mortality in low-income Black, high-income Black, and low-income White communities compared to high-income White communities [29]. A similar effect was seen from an extensive analysis using the US National Vital Statistics System, which showed that every unit decrease in PM2.5 levels would decrease associated mortality more in Black adults compared to White adults [30]. ...

Disparities in air pollution attributable mortality in the US population by race/ethnicity and sociodemographic factors

Nature Medicine

... Therefore, the World Health Organization (WHO) calls for the promotion of clean alternatives, mainly liquefied petroleum gas (LPG). Following this logic, LPG subsidization has been pursued in several countries 4 . Without such policies, biomass cooking will probably remain high in SSA for the decades to come 5 . ...

In praise of cooking gas subsidies: transitional fuels to advance health and equity * Our title draws inspiration from Kirk R. Smith (2002) ‘In praise of petroleum?’ Science and Kirk R. Smith (2014) ‘In praise of power’ Science.

... However, the heat maps also reveal that the model begins to underpredict more severely for observed PM 2.5 concentrations exceeding ≈ 200 μg/m 3 . The PM 2.5 -GNN model's R 2 value is lower than those reported in Considine et al. (2023) and Qiu et al. (2024), which both studied PM 2.5 predictions using machine learning, because our predictions are at an hourly resolution, whereas theirs are at a daily resolution. ...

Evaluating estimation methods for wildfire smoke and their implications for assessing health effects
  • Citing Preprint
  • June 2024