Increased extreme humid heat hazard faced by agricultural workers

Increases in population exposure to humid heat extremes in agriculturally-dependent areas of the world highlights the importance of understanding how the location and timing of humid heat extremes intersects with labor-intensive agricultural activities. Agricultural workers are acutely vulnerable to heat-related health and productivity impacts as a result of the outdoor and physical nature of their work and by compounding socio-economic factors. Here, we identify the regions, crops, and seasons when agricultural workers experience the highest hazard from extreme humid heat. Using daily maximum wet-bulb temperature data, and region-specific agricultural calendars and cropland area for 12 crops, we quantify the number of extreme humid heat days during the planting and harvesting seasons for each crop between 1979–2019. We find that rice, an extremely labor-intensive crop, and maize croplands experienced the greatest exposure to dangerous humid heat (integrating cropland area exposed to >27 °C wet-bulb temperatures), with 2001–2019 mean rice and maize cropland exposure increasing 1.8 and 1.9 times the 1979–2000 mean exposure, respectively. Crops in socio-economically vulnerable regions, including Southeast Asia, equatorial South America, the Indo-Gangetic Basin, coastal Mexico, and the northern coast of the Gulf of Guinea, experience the most frequent exposure to these extremes, in certain areas exceeding 60 extreme humid heat days per year when crops are being cultivated. They also experience higher trends relative to other world regions, with certain areas exceeding a 15 day per decade increase in extreme humid heat days. Our crop and location-specific analysis of extreme humid heat hazards during labor-intensive agricultural seasons can inform the design of policies and efforts to reduce the adverse health and productivity impacts on this vulnerable population that is crucial to the global food system.


Introduction
The effects of climate change are not distributed uniformly across the globe, and as warming continues, certain regions and communities will experience disproportionately more hazards and impacts (Ranasinghe et al 2021, U.S. Environmental Protection Agency 2021, Sun et al 2015, Forzieri et al 2018, Wang et al 2020).Among these changing hazards is that posed by extreme humid heat, which is the combined effect of high dry-bulb air temperature and high humidity and is often quantified by wet-bulb temperature (T W ). The frequency of humid heat extremes has increased substantially in the past four decades over densely-populated and socioeconomically vulnerable regions of the world at a faster rate than dry heat extremes, with a person on average experiencing about five more extreme humid heat days each decade (Rogers et al 2021).Future population exposure to T W exceeding 32°C, a threshold indicating potential adverse human health effects, is projected to increase 5 to 10-fold by the 2070s under different emissions scenarios due to dangerous humid heat becoming much more frequent (Coffel et al 2017), especially in densely-populated regions where current humid heat levels are already high (Rogers et al 2021).These increases will be especially large in the tropics (Wang et al 2020, Rogers et al 2021) and are also projected to last longer, and increase in intensity under global warming (Wang et al 2020).
Increases in T W pose a substantial risk to human health, yet historical changes and their impacts are not well understood.When combined with high dry-bulb temperatures, high specific humidity decreases the bodys ability to cool by sweating, exacerbating heat stress (Osilla et al 2022).Sherwood and Huber (2010) suggest a theoretical lethal T W limit of 35 °C based on a thermodynamic gradient from the core human body temperature to the ambient environment.However, recent experimental work by Vecellio et al (2021) suggests that this limit varies based on local climatological conditions and is potentially much lower than 35 °C T W .At T W of approximately 27 °C, heat stress shifts from demanding extreme caution to being dangerous for human health (Kang and Eltahir 2018 [table S6];National Weather Service, n.d.).Additionally, Hanna and Tait (2015) note that while humans have the capability to acclimatize to increasingly hot environments, climate change will introduce extreme T W conditions in which even highly acclimatized individuals, like outdoor workers, will not be able to sustain physical activity outdoors.
Agricultural workers are a population especially vulnerable to a proliferation of extreme humid heat events as they work outdoors for extended periods and often undertake strenuous physical labor.They are also among the most socio-economically vulnerable populations (Ergon Associates 2016).Frameworks have been developed that contextualize agricultural workers' risk of heat-related illness by examining the specific environmental exposures they face, the details of their physiological response and sensitivity to heat, possible adaptive measures that can reduce risk, as well as the sociological factors that increase their vulnerability (Mac and McCauley 2017).Riley et al (2018) have shown that United States counties with higher proportions of outdoor workers see more emergency room visits and hospitalizations during heat wave events than other counties.Similarly, recent work shows that chronic kidney disease in Central America has been linked to heat stress experienced while harvesting sugarcane (Wesseling et al 2020).Additionally, Spector et al (2016) found that agricultural workers in Washington state have a higher probability of traumatic injury during hot and humid days, especially during the summer harvest season.Recent modeling shows that even under moderate warming scenarios, heat risks to US agricultural workers are projected to increase, along with the number of days in agricultural regions that are classified as unsafe for outdoor work (Tigchelaar et al 2020).Consequently, modeling by de Lima et al (2021) projected large reductions in the productivity of agricultural workers in some regions, with limited potential for offsets via adaptation strategies, such as changing the time of day of agricultural labor.
Given the substantial impacts extreme humid heat poses to the health and productivity of agricultural workers and the global food system, it is imperative that we better understand the risks that extreme T W poses to these vulnerable populations around the world.Our study builds upon the limited research to date that links humid heat extremes to the heat stress hazards faced by agricultural workers, integrating an understanding of physiologically hazardous heat with data on the timing of agriculture across the world.First, we compute a global climatology of dangerous humid heat exposure over land, then, using crop-specific and region-specific agricultural calendars, we examine cropland exposure to humid heat during periods of the agricultural season when workers are likely to be in the field performing strenuous activities.Cropland exposure during laborintensive agricultural seasons is used as a proxy for agricultural worker exposure, which we refer to as potential worker exposure.Next, focusing on rice and maize, we analyze the seasonality (planting or harvest season) of this exposure and consider annual trends in overall labor-intensive season exposure over the study period.Lastly, we examine the impact of the El Niño Southern Oscillation (ENSO) on dangerous humid heat extremes over land in order to consider its role on potential worker exposure.Our work makes novel contributions to understanding the spatial and temporal distribution of humid heat hazards to agricultural workers across the world by linking the timing and location of agricultural activity with extreme daily T W using both absolute (T W > 27 °C) and relative (local 95th percentile of T W ) thresholds for 12 different crops.

Data and methods
We use three primary datasets in this analysis: (1) daily maximum T W data produced by Rogers et al (2021) using ERA5 reanalysis data (Hersbach et al 2020); the authors used the Davies-Jones method (Davies-Jones 2008) as implemented by Kopp (2020) and Buzan et al (2015) to calculate T W ; (2) the Global Crop Calendar Dataset -a gridded dataset of planting and harvest season dates for major global crops (Sacks et al 2010); and (3) harvested area from Monfreda et al (2008).The agricultural season dates were collected as benchmarks in the 1990s and early 2000s (Sacks et al 2010) while the harvested area data were estimated for the year 2000 (Monfreda et al 2008); these data are thus time invariant for this analysis.
We analyze the occurrence of extreme humid heat days during the crop-specific planting and harvest seasons, both individually and combined (hereafter referred to as the labor season), using the planting and harvest season dates from the Global Crop Calendar Dataset over the 1979-2019 period.This dataset provides beginning and ending planting and harvest season dates for each crop in each grid cell where the crop is harvested, these delimit the seasons for our analysis.An example of the planting and harvest seasons for rice and maize as tabulated in the Global Crop Calendar Dataset is shown in figure 1.For most of the regions, planting season for both crops (figures 1(a), (b)) starts in late winter and early spring, and harvest season (figures 1(c), (d)) falls between late summer and autumn.The labor seasons shown in figures 1(e), (f) tend to last four or more months for most rice and maize growing regions.The reasons for the papers focus on rice and maize are explained in Results section 3.2.
Two thresholds are used to define extremes: an absolute threshold of 27 °C daily maximum T W (T W c 27 ), a threshold considered to be dangerous to outdoor workers (Kang and Eltahir 2018, National Weather Service, n.d.), and a relative threshold defined as the local (grid cell) 95th percentile of daily maximum T W over 1979-2019 (T W p 95 ).The inclusion of the relative threshold allows us to investigate local humid heat extremes in relatively cooler regions that do not often exceed T W c 27 , and hence could pose health risks to communities not acclimatized to such conditions.It also allows us to investigate extreme conditions in regions where humid heat extremes are often higher than T W c 27 .The total number of days exceeding T W c 27 and T W p 95 are calculated for each year over each land grid cell of the Global Crop Calendar Dataset (0.5°× 0.5°), which requires a regridding of the 0.28125°× 0.28125°ERA5 data.Extreme humid heat days for each grid cell are then aggregated to their corresponding crop-specific agricultural season.Grid cells that do not overlap with the harvested area cells of the crop from Monfreda et al (2008) are masked.This procedure is similar to that of Gourdji et al (2013), who investigated exceedances of dry-bulb temperatures dangerous to certain crops during their reproductive seasons.

Climatology and trends in humid heat extremes
We first examine global land regions where extreme humid heat events, as defined by our physiologically-based thresholds, tend to occur frequently and identify trends over the past several decades (figure 2). Figure 2(a) reveals that some land areas experience three or more months of T W temperatures exceeding 27 °C (> T W c 27 ) per year on average.These regions include the Amazon, the Indo-Gangetic Basin, Northern Colombia, the Mexican coast, the coasts of the Red Sea and Persian Gulf, Southeast Asia, and the Maritime continent.The northern coast of the Gulf of Guinea, Senegal and the Gambia, and the northern coast of Australia also experience approximately two months of extreme humid heat.While the majority of these locations are in the tropics and are near-coastal, there are several sub-tropical and inland exceptions, such as northern South Asia, and northeastern and southeastern China.surprisingly, exhibit a similar pattern (figure 2(d)) despite by definition featuring the same number of days everywhere in the climatology.The largest increases occur in some of the most densely populated regions in the world, such as the Mekong Delta, and Bangladesh/West Bengal, with large increases also found in parts of the Amazon, northern Colombia, and the Maritime Continent.T W p 95 exceedance trends are stronger and extend over a greater geographic region, with relatively weak yet significant trends extending to cooler mid-and highlatitude land regions.Overall, the consistency of the exceedance trends for both thresholds reveals increases in extreme T W exposure over many land regions.In contrast, very few regions see decreases in extreme T W exposure (e.g.parts of northern Australia, figure 2(d)).Figure 2 is restricted to three regions of interest, Central/ South America, South Asia and the Maritime Continent, and the Gulf of Guinea in figures A1, A2, and A3.

Integrated potential worker exposure to humid heat during the agricultural labor season
To identify crops whose workers are subject to the greatest potential exposure to humid heat during the laborintensive seasons (figure 1), we multiplied the cropland area in each grid cell by the number of extreme T W days exceeding the threshold values for each crops season, then integrated this across all grid cells (figure 3).We use this cropland exposure as a proxy for potential exposure of crop workers, as there are no comprehensive data on the times of day various crop workers are in the field.All original 25 crops in the Global Crop Calendar Dataset were analyzed with this metric; the 12 crops with the highest consistent values across both thresholds are shown in figure 3.This aggregated measure of exposure depicts the effect of both the extent of area exposed and the frequency of humid heat exposure for each crop across the globe, with the assumption of constant cropland area from 1979-2019.This likely leads to an underestimate, given the expansion of global cropland in the 21st century, notably in South America (Potapov et al 2021).Using T W c 27 , rice, maize, sorghum, and soybeans are the crops whose workers are the most potentially exposed to extreme T W (figure 3(a)).We also examined potential worker exposure using T W p 95 to capture potential impacts from locally extreme conditions (figure 3 .This discrepancy occurs because maize is more common than rice in temperate areas that rarely exceed T W c 27 (figures 1(e), (f) and 2(a)).For both thresholds there is a strong ENSO signal in potential worker exposure, particularly for 1998, 2010, and 2016 (figure 3).This reflects the ENSO signal in humid heat frequencies (figure 2(b)).Because figure 3 indicates that rice and maize agricultural workers are the most potentially exposed, we focus the remainder of our detailed analysis on these two crops.Potential worker exposure for rice and maize nearly doubled during the study period, with the mean for the latter half of the period (2001-2019) being 1.8 times (rice) and 1.9 times (maize) greater than those for the earlier period .For rice, this is a 1979-2000 mean of ∼1.08 hundred million area days and a 2001-2019 mean of ∼1.94 hundred million area days.

Potential worker exposure for rice and maize
Given their large increases in potential exposure (figure 3), we examine the spatial and temporal distribution of humid heat extremes for rice and maize.Both rice and maize are widely cultivated, with rice dominant in Asia, South America, and central Africa, whereas maize is prevalent in North America, southern South America, and southern Africa.Rice cropland in this dataset covers 35.8 million square kilometers and maize 43.5 million square kilometers.As shown in figure 1, planting of both rice and maize outside the tropics starts in late winter and early spring, and the harvest season starts ∼2-4 months later.Maize planting and harvest seasons for countries in Southeast Asia and the Maritime Continent differ more from their neighbors than those in the midlatitudes.For example, regions of Myanmar plant maize in September, while regions in Thailand plant in April and parts of Sumatra plant in November, with similarly disjointed harvest seasons (figures 1(a), (b), (c), (d)).The total labor season for rice and maize is generally less than ∼7 months (figures 1(e), (f)).Regions with notably long labor seasons covering most of the year are all in the tropics, including Cuba and French Guiana (rice, figure 1(e)), and Venezuela and eastern Australia (maize, figure 1(f)).
Figure 4 shows the potential worker exposure to extreme humid heat for rice and maize in terms of the average days exceeding the threshold values (figures 4(a), (b)), their trend over the last 4 decades (figures 4(c), (d)), and the relative potential exposure in the planting versus the harvest seasons (figures 4(e), (f)).(Figure 4 is restricted to the three regions of interest, Central/South America, South Asia and the Maritime Continent, and the Gulf of Guinea in figures A4, A5, and A6).For rice, the highest potential worker exposure during the labor season occurs in Bangladesh, with more than 60 days of extreme humid heat >T W c 27 (figure 4 A1).For rice, 10.1 million square kilometers of cropland are more exposed in the harvest season, compared to 7.8 million square kilometers more exposed in the planting season.These exposures represent 30.5% and 22.3% of total rice cropland, respectively.For maize, 12.2 million square kilometers are more exposed in the harvest season than in the planting season, and 5.7 millions square kilometers are more exposed in the planting season than the harvest season, representing 28.1% and 13.1% of total maize cropland.Regions where potential exposure is biased towards one season over the other generally show a strong bias in that season.Considering areas where exposure is strongly biased towards a season (where there are at least twice as many mean exposures in one season over the otherratio <0.5 for strong harvest exposure or ratio >2 for strong planting exposure) ∼22% of both croplands experience strongly biased exposure in the harvest season, which is notably large considering ∼30% of the croplands are generally more exposed (ratio <1) in the harvest season.Similarly, 16.6% percent of rice cropland exposure is strongly biased towards the planting season (compared to 22.3% generally more exposed, ratio >1, in the planting season), and 9.2% of maize cropland exposure is strongly biased towards the planting season (compared to 13.1% generally more exposed in the planting season).
More rice cropland is exposed in the planting season compared to maize.About 30% of both croplands are more exposed in the harvest season, but almost twice as much rice cropland is more exposed in the planting season, by percentage of total cropland, than maize.In absolute terms, rice has ∼2 million square kilometers more cropland more exposed in the planting season than maize.Considering strongly biased season exposure, about twice as much rice cropland exposure is strongly biased towards the planting season than is for maize cropland.

Humid heat extremes and the El Niño-southern oscillation
Given the high humid heat frequencies (Figure 2(b)) and high potential worker exposure (Figure 3) during El Niño years, we examine the frequency of annual extreme T W over the nine moderate to very strong (ONI >1) El Niño events (1983, 1987, 1988, 1992, 1995, 1998, 2003, 2010, and 2016) as compared to all other years (figure 6).The years examined here are the latter years of two-year El Niño events, where figure 3 indicates the increased exposure signal.To remove the secular warming trend, extreme humid heat days have been linearly detrended before compositing the El Niño years.El Niño events are associated with enhanced T W c 27 exceedances over Southeast Asia, Sumatra, the Amazon, southern Pakistan, Bangladesh, and the northern coast of the Gulf of Guinea (figure 5(a)).These regions also have the highest mean humid heat frequencies for all years (figure 2(a)) Thus, our analysis indicates that during an El Niño event, many agricultural regions are likely to experience higher exposure to T W extremes, exacerbating the already dangerous T W conditions that many agricultural workers already face.

Summary and discussion
In this study, we characterize the climatology and trends in potential agricultural worker exposure to hazardous humid heat conditions across regions for various key crops based on the crop-specific timing of planting and harvest activity.Increases in extreme humid heat frequencies over croplands during the agricultural labor seasons imply that agricultural workers will be more at risk from these events.Our use of crop and locationspecific agricultural season calendars highlights large differences in humid heat exposure across crops and locations.Consistent with prior historical analyses (Raymond et al 2020, Rogers et al 2021) and projections (Coffel et al 2017) over the calendar year, we find the highest baseline frequency and trends in humid heat days  .
occur in the tropics and lower subtropics.We find that for many crops, the area extent and frequency of potential agricultural worker exposure is increasing rapidly, particularly in the past ∼15 years (figure 3).
Rice, maize, sorghum and soybean croplands are the most exposed to dangerous humid heat, and thus, in this analysis, represent the most hazardous croplands where agricultural workers may face dangerous T W conditions.In particular, rice and maize planting seasons in Southeast Asia, coastal Mexico, the northern coast of the Gulf of Guinea, and Pakistan experience greater impacts; whereas, impacts are higher during the harvest seasons in South America, eastern China, and the Maritime Continent.We also find that El Niño conditions dramatically increase cropland exposure to dangerous humid heat, particularly since the 1997/98 El Niño.The regions of increased dangerous humid heat frequencies during El Niño years also have high climatological frequencies and trends.This emphasizes the need to better understand the influence of natural climate variability modes and their interactions with the mean warming trends to characterize and predict threats to vulnerable agricultural workers that are a backbone of the global food system.
Further, there are several fertile avenues for further research on impacts and potential adaptation strategies.First, the actual exposure and vulnerability of agricultural workers may differ dramatically from our first-order estimates based on a single humid heat metric and two extreme thresholds.The compound effects to agricultural workers from exposure to consecutive humid heat extremes during heat waves, for example, is one critical component not examined in this work.Second, our assumption that crop calendars, crop density, and worker distributions are stationary and spatially homogeneous below the regional scale shown in figure 1 also limits the generalizability of our results, as they are all dynamic variables.Future research should thus explore factors including the intensity of the labor, the number of workers, their degree of vulnerability (based on factors including pre-existing health conditions, gender, age, access to cooling for recovery when not farming, access to technological innovations such as air-conditioned tractors, work compensation structure such as piece-rate work, and local labor rights and protections), time of day of exposure to humid heat, shifting crop calendars, the regional proliferation and recession of agriculture, and factors such as solar radiation, surface winds, and exposure to humid heat thresholds higher than 27 °C T W . Efforts toward a comprehensive census of agricultural workers would greatly improve our ability to refine the details of humid heat hazards to these vulnerable populations and could help better inform public policy aimed at mitigating health risk.Third, additional research is needed to explore the indirect impacts of humid heat on crop productivity via its impacts on agricultural worker health and productivity.While inherently important, this human health and productivity information could also be integrated with direct impacts of climate change on crops (Iizumi andRamankutty 2016, Deryng et al 2014).For heat specifically, direct decreases in crop productivity are found to mostly occur due to extreme dry heat (Ting et al 2023) associated with high dry-bulb temperatures.While it is tempting to speculate that a tendency towards humid heat might lessen negative yield impacts, the humid heat impacts on labor productivity through health impact on workers could in principle be as impactful as direct plant impacts to overall crop productivity.Additional empirical research (along the lines of de Lima et al 2021) and modeling are needed to fully understand the relationship between losses in labor productivity and crop yield impacts.The potential for deleterious feedback on food security between direct damage to crops and workers in a warming world enhances the urgency of this research.Fourth, ours and others findings hint at proximity to water, whether warm water bodies and coastlines (Raymond et al 2022), or river valleys and irrigated land (Monteiro and Caballero 2019, Krakauer et al 2020, Thiery et al 2017) as drivers of humid heat.Since agriculture also often leverages this same proximity to water, future research should explore whether proximity of croplands to water influences worker humid heat exposure in ways not captured here.Indeed, the intensification and expansion of  (1983, 1987, 1988, 1992, 1995, 1998, 2003, 2010, and 2016).Composites are calculated after removing background linear trends in extreme humid heat day frequencies.Composites are produced using data for the entire calendar year.
agriculture in water-rich regions like the Amazon (Simon and Garagorry 2005) may already be increasing both the frequency of, and agricultural worker exposure to, dangerous humid heat, as has been suggested for the Indo-Gangetic and Indus Basins (Mishra et al 2020, Krakauer et al 2020).This possibility is hinted at by our findings of strong trends in humid heat days in regions like the Amazon and Southeast Asia, but more research is needed.The possibility that irrigation may change the timing of agricultural activity, and thus the seasons in which crop workers are engaged in outdoor work, must also be considered in future work.
Our research clearly shows that climate change, interacting with modes of natural variability such as ENSO, is already increasing the exposure of agricultural croplands, and thus highly vulnerable agricultural workers, to extreme humid heat.Not only are these workers essential to the global food system, but their livelihoods are threatened in the face of such increasing health hazards.Our initial effort here at identifying crop and locationspecific potential agricultural worker exposure to extreme humid heat is a necessary early step in the long-term effort to help workers and the agricultural systems they support adapt to climate change, climate variability, and other compound stresses (Raymond et al 2020) in a rapidly changing world.Table A1.Breakdown of Global Rice and Maize Cropland Exposure by Season.'Ratio' refers to the ratio of mean extreme humid heat days (>T W c 27 ) in the planting to harvest seasons for the given crop.Exposed cropland is based on the climatological mean potential worker exposure calculated over the study period (1979-2019).

Rice Maize
Total Cropland (km       (1983, 1987, 1988, 1992, 1998, 2003, 2010, and 2016).Composites are calculated after removing background linear trends in extreme humid heat day frequencies.Composites are produced using data for the entire calendar year.(1989, 1996, 1999, 2000, 2008, 2011, and 2012).La Niña years are considered the latter year of a two-year event, as for El Niño events analyzed in the main text.Composites are calculated after removing background linear trends in extreme humid heat day frequencies.Composites are produced using data for the entire calendar year.

Figure 1 .
Figure 1.Rice and maize planting and harvest seasons.Month in which the first day of (a)-(b) planting and (c)-(d) harvest occurs.(e)-(f) Length of the agricultural labor season (combined planting and harvest seasons).Season data spatial extent is masked according to the coverage of the harvested area data from Monfreda et al (2008).
Averaged over global land areas, exceedances of T W c 27 more than doubled over 1979-2019, from ∼2-4 days of extreme humid heat per year in the 1980s to ∼4-8.5 days in the 2010s (figure 2(b)), an increase of ∼1 day per decade.In comparison, global exceedances of the local 95th percentile show a stronger increase of ∼3 days per decade.Both of these trends are statistically significant as described in figure 2. Both extreme humid heat indices show peaks during strong El Niño years, particularly 1998, 2010 and 2016.While the 1983 and 1987 El Niño events also show a relatively high number of extreme humid heat days, they are less marked relative to the surrounding non-El Niño years.The differing influence of various El Niño events could be due to a number of factors, including duration, timing, peak strength, and interactions with the mean warming trend.Many regions with a climatologically high number of extreme T W days per year (>T W c 27 , figure 2(a)) also show large positive trends in exceedances of this metric (figure 2(c)).Trends in T W p 95 exceedances, more

Figure 2 .
Figure 2. Climatology and trends in extreme humid heat frequencies.(a) Number of days per year exceeding T W c 27 over land; note that the color bar saturates at 93 days (∼3 months).(b) Area-weighted average T W p 95 exceedances (yellow x markers, right y-axis) and T W c 27 exceedances (blue + markers, left y-axis) Note the different vertical axes for T W c 27 and T W p 95 exceedances.Averages are calculated over land only.Trends are significant at the 95% confidence level using the Mann-Kendall test (p-values <0.001).Dotted blue and dashed yellow lines show linear trendlines for T W c 27 and T W p 95 , respectively.Linear trends in annual days exceeding (c) T W c 27 and (d) T W p 95 .Hatching indicates trends that are not significant at the 95% confidence level using the Mann-Kendall test (p-values <0.001).
(b)).While rice, maize, and sorghum have the greatest potential worker exposures to humid heat for T W p 95 (figure 3(b)), consistent with T W c 27 , soybean exposure is lower relative to other crops for T W p 95 than for T W c 27 .Additionally, maize exposure for T W p 95 exceeds that for rice, whereas the opposite is true for T W c 27

Figure 3 .
Figure 3. Integrated potential worker exposure to extreme humid heat.Cropland area in each grid-cell (km 2 ) multiplied by the number of days in the agricultural labor season exceeding (a) T W c 27 and (b) T W p 95 for each grid-cell.Area days are in hundred million square kilometer days.
(a)).Other regions with relatively high frequencies include the Mekong Delta, coastal Thailand, the Irrawaddy River Delta in Myanmar, the Maritime Continent, and parts of coastal Mexico and the Amazon, which show 30 or more days of extreme humid heat >T W c 27 (figure 4(a)).Positive trends in potential worker exposure (figure 4(c)) are observed where extreme heat frequencies are already high for rice, with strong trends occurring in Southeast Asia and parts of the Amazon (10 days per decade).For the maize labor season, the highest potential worker exposure occurs across much of Pakistan and parts of Mekong Delta (figure 4(b)).Other regions with high potential worker exposure include northern Colombia and Venezuela, the Philippines, parts of coastal Mexico, and coastal Iran (figure 4(b)).The strongest positive trends are found in northern Colombia, Venezuela, and the Mekong Delta, which also have high humid heat frequencies similar to the rice frequency trends at 10 days per decade (figures 4(b), (d)).Figures 4(e)-(f) indicates regional patterns in which season more potential worker exposure occurs.In general, Southeast Asia, India, Pakistan, and the northern coast of the Gulf of Guinea are more exposed in the planting season for both rice and maize.eastern China and the Maritime Continent are generally more exposed in the harvest season for both crops.Regional patterns are less clear for South America, Sub-Saharan Africa, Mexico, and the southern United States where the season of more frequent potential worker exposure depends on the specific crop, and shows more small-scale spatial variability.These patterns are also shown in the frequency, trend, and seasonal bias maps for Sorghum in figure A7.These seasonal findings are consistent with the peak timing of humid heat extremes in Rogers et al (2021).For example, the authors identify May as having the highest frequency of extreme T W p 95 days over Cambodia and Thailand, while we similarly identify this month as the start of the season with the most frequent T W c 27 exposure.Area calculations from figures 4(e)-(f) shown in figure 5 also indicate that extreme humid heat exposure tends to occur more frequently during the harvest season (total area values in km 2 are tabulated in table

Figure 4 .
Figure 4. Spatial distribution of potential worker exposure in the labor season for rice and maize.Mean number of extreme humid heat days during the labor season exceeding the absolute threshold T W c 27 for (a) rice and (b) maize.Linear trends in the number of labor season days per decade exceeding T W c 27 for (c) rice and (d) maize.Hatching shows where trends are not significant using the Mann-Kendall test at the 95% confidence level (p-values < 0.001).Ratio of the mean number of humid heat days exceeding T W c 27 in the planting season to those in the harvest season for (e) rice and (f) maize.Values >1 (green, right) show more potential worker exposure in the planting season, values <1 (purple, left) show more potential worker exposure in the harvest season.Areas with no humid heat days in either season are masked in dark grey; areas with no humid heat days in the harvest season, but 1 day in the planting season are reported as a ratio of >2, where the colorbar saturates.

Figure 5 .
Figure 5. Breakdown of Global Rice and Maize Cropland Exposure by Season.Proportions of the total cropland exposed in the indicated season are shown.Exposed cropland is based on the climatological mean potential worker exposure calculated over the study period (1979-2019).

Figure 6 .
Figure 6.Mean humid heat frequency anomalies during El Niño events.Composites of the anomalous annual number of extreme humid heat days for (a) T W c 27 and (b) T W p 95 , during years with moderate to very strong ONI El Niño events(1983, 1987, 1988, 1992,  1995, 1998, 2003, 2010, and 2016).Composites are calculated after removing background linear trends in extreme humid heat day frequencies.Composites are produced using data for the entire calendar year.

Figure A1 .
Figure A1.Climatology and trends in extreme humid heat frequencies over Central and South America.(a) Number of days per year exceeding T W c 27 over land; note that the color bar saturates at 93 days (∼3 months).Linear trends in the number of days per decade exceeding (b) T W c 27 and (c) T W p 95 .Stippling indicates trends that are not significant at the 95% confidence level using the Mann-Kendall test (p-values <0.001).

Figure A2 .
Figure A2.Climatology and trends in extreme humid heat frequencies over South Asia and the Maritime Continent.(a) Number of days per year exceeding T W c 27 over land; note that the color bar saturates at 93 days (∼3 months).Linear trends in the number of days per decade exceeding (b) T W c 27 and (c) T W p 95 .Stippling indicates trends that are not significant at the 95% confidence level using the Mann-Kendall test (p-values <0.001).

Figure A3 .
Figure A3.Climatology and trends in extreme humid heat frequencies over the Gulf of Guinea (a) Number of days per year exceeding T W c 27 over land; note that the color bar saturates at 93 days (∼3 months).Linear trends in the number of days per decade exceeding (b) T W c 27 and (c) T W p 95 .Stippling indicates trends that are not significant at the 95% confidence level using the Mann-Kendall test (p-values <0.001).

Figure A5 .
Figure A5.Spatial distribution of potential worker exposure in the labor season for rice and maize in South Asia and the Maritime Continent.Mean number of extreme humid heat days during the labor season exceeding the absolute threshold T W c 27 for (a) rice and (b) maize.Linear trends in the of labor season days per decade exceeding T W c 27 for (c) rice and (d) maize.Stippling shows where trends are not significant using the Mann-Kendall test at the 95% confidence level (p-values < 0.001).Ratio of the mean number of humid heat days exceeding T W c 27 in the planting season to those in the harvest season for (e) rice and (f) maize.Values >1 (green, right) show more potential worker exposure in the planting season, values <1 (purple, left) show more potential worker exposure in the harvest season.Areas with no humid heat days in either season are masked in dark grey; areas with no humid heat days in the harvest season, but 1 day in the planting season are reported as a ratio of >2, where the colorbar saturates.

Figure A6 .
Figure A6.Spatial distribution of potential worker exposure in the labor season for rice and maize in the Gulf of Guinea.Mean number of extreme humid heat days during the labor season exceeding the absolute threshold T c 27 for (a) rice and (b) maize.Linear trends in the number of labor season days per decade exceeding T W c 27 for (c) rice and (d) maize.Stippling shows where trends are not significant using the Mann-Kendall test at the 95% confidence level (p-values < 0.001).Ratio of the mean number of humid heat days exceeding T W c 27 in the planting season to those in the harvest season for (e) rice and (f) maize.Values >1 (green, right) show more potential worker exposure in the planting season, values <1 (purple, left) show more potential worker exposure in the harvest season.Areas with no humid heat days in either season are masked in dark grey; areas with no humid heat days in the harvest season, but 1 day in the planting season are reported as a ratio of >2, where the colorbar saturates.

Figure A8 .
Figure A8.Mean humid heat frequency anomalies during El Niño events over rice and maize cropland.Composites of the anomalous annual number of extreme humid heat days for T W c 27 over (a) rice cropland (c) maize cropland and T W p 95 over (b) rice cropland and (d) maize cropland, during years with moderate to very strong ONI El Niño events(1983, 1987, 1988, 1992, 1998,  2003, 2010, and 2016).Composites are calculated after removing background linear trends in extreme humid heat day frequencies.Composites are produced using data for the entire calendar year.

Figure A7 .
Figure A7.Spatial distribution of potential worker exposure in the labor season for sorghum.(a) Mean number of extreme humid heat days during the labor season exceeding the absolute threshold T W c 27 .(b) Linear trends in the number of labor season days per decade exceeding T W c 27 .Hatching shows where trends are not significant using the Mann-Kendall test at the 95% confidence level (pvalues < 0.001).(c) Ratio of the mean number of humid heat days exceeding T W c 27 in the planting season to those in the harvest season.Values >1 (green, right) show more potential worker exposure in the planting season, values <1 (purple, left) show more potential worker exposure in the harvest season.Areas with no humid heat days in either season are masked in dark grey; areas with no humid heat days in the harvest season, but 1 day in the planting season are reported as a ratio of >2, where the colorbar saturates.

Figure A9 .
Figure A9.Mean humid heat frequency anomalies during La Niña events.Composites of the anomalous annual number of extreme humid heat days for (a) T W c 27 and (b) T W p 95, during years with moderate to very strong ONI La Niña events(1989, 1996, 1999, 2000,  2008, 2011, and 2012).La Niña years are considered the latter year of a two-year event, as for El Niño events analyzed in the main text.Composites are calculated after removing background linear trends in extreme humid heat day frequencies.Composites are produced using data for the entire calendar year.