1

Tropical deforestation already outpacing climate change heat impacts, limiting human ability to safely work outside

Luke A. Parsons*1,2, Jihoon Jung3, Yuta J. Masuda4, Lucas R. Vargas

Zeppetello1, Nicholas H. Wolff5, Timm Kroeger4, David S. Battisti1, June T. Spector3

1 Department of Atmospheric Sciences, University of Washington, Seattle, WA 2 Nicholas School of the Environment, Duke University, Durham, NC 27708 3 Departments of Environmental and Occupational Health Sciences and Medicine, University of Washington, Seattle, WA 4 Global Science, The Nature Conservancy, Arlington, VA 5 Global Science, The Nature Conservancy, Brunswick, ME * Corresponding Author: Luke A Parsons: luke.parsons@duke.edu

Abstract

Almost one-fifth of tropical forest cover has been lost in the last two decades, leading to local temperature increases that can surpass warming from 21st century climate change projections. Although it is known that working outdoors in high temperatures reduces worker productivity and increases heat strain, the extent to which deforestation-driven temperature change affects people across the tropics is unknown. Using satellite data combined with worker health guidelines, we show that warming associated with deforestation is already impacting human health and well-being in low latitude countries. We estimate that more than 1.4 million people in recently deforested tropical biome now face heat exposure that reduces safe work hours by at least two hours per day. This warming has particularly large impacts on populations in Brazil, Belize, Cambodia, Vietnam, Malaysia, Myanmar, Nigeria, and Cameroon. We highlight these effects by examining the Brazilian states of Mato Grosso and Pará, which have experienced particularly large-scale deforestation; we show that temperatures and human heat exposure are increasing disproportionately quickly in these locations. Future global climate change magnifies heat exposure in deforested areas across the tropics. An additional 2°C of global warming will increase exposure to multiple hours of unsafe working conditions for over 9.5 million people in recently deforested areas. Tropical deforestation is hastening the arrival of climate change impacts in the tropics, highlighting the need to shift local land use practices and to slow global greenhouse gas emissions so the most vulnerable populations are not forced to bear the brunt of warming impacts.

2

Main Text Tropical forests serve as global carbon sinks and provide critical ecosystem services for local communities (1), but tropical deforestation driven by the expansion of agriculture and logging has accelerated in recent decades (2). Trees help regulate the local thermal environment (3,4). The loss of tropical forests removes these important cooling services and leads to increased temperatures (Ellison et al., 2017; McAlpine et al., 2018; Wolff et al., 2018; Cohn et al., 2019; Masuda et al., 2019), mediated by changes in shade, evapotranspiration, and surface reflectivity (Prevedello et al., 2019). Tropical deforestation often occurs in contiguous patches of land that are cleared to create agriculture or pasture land. The magnitude of local warming in these patches is strongly associated with their area, with annually averaged maximum temperatures that are up to 10°C warmer in large (>100 km2) deforested patches (5) and warming effects that can extend up to 50 km beyond deforested sites (6). Local warming of this magnitude may be accelerating the onset of heat exposure of up to 3 billion people who are expected to exit the human climate niche in the next 50 years (7).

Deforestation-induced loss of cooling services may be particularly detrimental to the health and well-being of communities in the low latitudes, as they are often dependent on outdoor work and have comparatively low adaptive capacity to adjust to environmental change (8, 9). Increased exposure to higher temperatures can threaten the resilience of communities, particularly if temperatures in their environment regularly exceed thresholds for human safety (10,11). Contributors to heat stress include ambient exposure to high temperature and humidity, internal heat generated by heavy physical work, and non-breathable clothing (12). Working in hot environments can increase core body temperatures, leading to heat strain and potentially fatal heat stroke, even among the young and otherwise healthy. Although it is well established that working outdoors in high temperatures reduces productivity and increases heat strain (13), the extent to which deforestation-driven temperature change affects human populations across the tropics is unknown. This is noteworthy because high temperature and humidity already create mid-day unsafe working conditions across much of the tropics (14).

Studies on deforestation and temperature changes have focused on quantifying warming from tropical deforestation at regional scales (5,15), and the relationship between deforestation and increasing temperatures in the tropics is well-established (Ellison et al., 2017; McAlpine et al., 2018; Wolff et al., 2018; Cohn et al., 2019; Masuda et al., 2019). However, only a few small-scale studies have evaluated deforestation effects on heat exposure and well-being of nearby populations in rural communities (16,17,18,19), limiting our understanding whether these effects extend globally in similar settings across low-latitude countries. Here, we show results from an analysis that quantifies the magnitude and geographic extent of deforestation impacts on changes in local temperature and associated heat exposure across 94 low-latitude countries with tropical forests. We estimate the safe work hours (the amount of time in a day during which heavy physical outdoor work can be performed safely under established heat exposure thresholds) that have been lost due to temperature increases from recent deforestation. We show where large proportions of populations are already exposed to unsafe working conditions due to heat exposure from tree cover loss in the last 15 years alone (2003-2018), identify areas under threat given future global warming, and discuss the implications of these findings for land use and outdoor workers in the tropics.

3

Results

Large-scale tropical deforestation and temperature increases Human activity has driven large-scale deforestation in the tropics. Of the ~23.07 million km2 encompassing the potential tropical and subtropical forest biome (green regions in Figure 1a, ref. 20), 13.9 million km2 was covered in >75% tree cover in the year 2000 (grey regions in Figure 1a, Methods, ref. 21). Between 2003 and 2018, 8.7% of this tropical forest cover (1.22 million km2) was lost or degraded (red regions in Figure 1a), with 6.9% forest cover loss in the Americas (~0.56 million km2), 6.5% loss in Africa (~0.14 million km2), and 13.6% loss in Asia (~0.52 million km2). Tropical deforestation is associated with local warming that exceeds interannual climate variability near the equator (5) and across the tropics and subtropics, particularly for afternoon temperatures analyzed here (Methods). Specifically, in the Americas, Africa, and Asia, low-latitude locations that maintained forest cover between 2003 and 2018 experienced mean temperature variations of ~0.2°C, ~0.4°C, and ~-0.01°C, respectively; these changes are consistent with climate model representations of internal climate variability and anthropogenic climate change. By contrast, deforested locations experienced mean increases in afternoon temperatures of 1.6°C, 1.2°C, and 0.55°C, respectively (Figure 1b; see also ref. Prevedello et al., 2019). The impacts of deforestation are even more apparent in the tails of the distributions showing the largest year on year temperature increases (5). For example, in the Americas, ~78% of locations that experienced afternoon temperature increases greater than 6°C were deforested between 2003 and 2018 despite these areas accounting for less than 9% of the original forested area. Although forest cover gain is associated with temperature decreases similar in magnitude to temperature increases from forest loss (Prevedello et al., 2019; Figure 1b), here we focus on deforestation due to data limitations associated with timing of tree regrowth (Methods). Decreases in safe work hours for heavy physical outdoor labor Our results indicate that temperature increases associated with deforestation have led to losses in safe work hours. Here ‘lost safe work hours’ is defined as the amount of time per day deemed unsafe for heavy physical outdoor work for given heat index values, which reflects heat exposure during working hours rather than individually-verified working hours (Methods). We find increased average lost safe work hours per day in deforested areas of 1.06 in the Americas, 0.54 in Africa, and 0.85 in Asia (Figure 1c). Large areas of land in the tropics experience some degree of year to year weather and climate variability and therefore some variation in lost safe work hours regardless of forest cover (Figure 1c), but a striking increase in heat exposure in recent decades has occurred in locations that have experienced deforestation (e.g., ref. Masuda et al., 2019). Between 2003 and 2018, the proportion of land areas experiencing more than 30 minutes of safe working time per day was higher in deforested compared to forested areas. For example, in the Americas, 5% of forested land areas experienced this magnitude of work hours lost, whereas 35% of deforested areas experienced the same loss (Figure 2a). Using Gridded Population of the World (GPWv4) data (22), we are able to approximate how many people are exposed to these increases in lost safe work hours (Methods). We find that during the analysis period, the proportion of the population experiencing increases of at least 30 minutes of lost safe work time per day was higher in deforested compared to forested areas. For example, in the American tropics, 1.6% of people living in forested areas experienced a loss of 0.5 safe work hours per day between 2003 and 2018, whereas 8.2% of people living in deforested areas experienced similar losses. We find similarly disproportionate lost work hours in Africa and Asia. In Africa, 8.8% of people living in deforested areas experience a loss of 0.5 safe work hours per day, whereas only 1.1% of people in forested areas experience a loss of 0.5 safe work hours per day. In Asia, 15.4% of people living in deforested areas experience a loss of 0.5

4

safe work hours per day, whereas 4.7% of people in forested areas experience a loss of 0.5 safe work hours per day (Figure 2b). Loss of more than 0.5 safe work hours per day during 2003-2018 is relatively common across the tropics in deforested areas, where large fractions of the working day were lost in some places, an effect of deforestation virtually non-existent in regions that retained their forest cover between 2003 and 2018 (Figure 1c; Figure 2). Over 1.4 million people already experience at least two hours of lost safe work hours per day in recently deforested areas in low latitude countries (Table S1). Increases of two lost safe work hours between 2003 and 2018 occurred in 6% of forested locations, and in 93.8% of deforested locations in the Americas. Human populations are highly concentrated in certain geographic areas, so regional and global analyses can mask locations with particularly large impacts of deforestation and temperature on populations. Therefore, we also examine the total number of people living in regions where more than 0.5 safe working hours per day was lost between 2003 and 2018 by country administrative division 1 - the highest subnational unit for each country (22). The size of the rings in Fig. 3 denotes the total number of people living in deforested areas by administrative division; the ring color denotes the percent of people in deforested areas who have lost at least 0.5 safe working hours per day between 2003 and 2018. Darker red colors in Figure 3 highlight ‘hotspots’, or administrative divisions with >30% of population in recently deforested areas, where the warming driven by deforestation impacts large portions of the population. We find ‘hotspots’ in Brazil, Belize, Cambodia, Vietnam, Malaysia, Myanmar, Nigeria, and Cameroon. Comparison of our predicted work loss impacts with Global Rural-Urban Mapping Project (GRUMP) urban areas and settlement points indicates that heat exposure in forest loss areas has increased both in rural and more densely populated areas. Increased deforestation-driven heat exposure in the latter areas is not surprising given that the rapid urban expansion observed in recent decades in many tropical countries has come partially, and in some cases, predominantly at the expense of forests (23). We have focused our analysis on heat exposure in the overall population, but by combining general population data with outdoor worker statistics (i.e., proportion of population working in agriculture, forestry, and fisheries sectors, ref. 24), we can also estimate the number of outdoor workers exposed to increases in lost safe work hours associated with deforestation in 41 countries where these data are available (Methods). We find that the proportion of outdoor workers that have lost more than 0.5 safe working hours per day between 2003-2018 is higher in deforested as compared to forested locations. In the Americas, Africa, and Asia, we find that 7.9%, 4.3%, and 10.8%, respectively, of outdoor workers in areas deforested during 2003-2018 lost at least 0.5 safe work hours per day due to forest loss, with more than 7.9 million outdoor workers exposed to 0.5 safe work hours per day in 2018 (Table S2). By contrast, only 1.24%, 0.16%, and 2.79% of outdoor workers in areas that maintained forest cover experienced similar losses in safe work hours. Global warming impacts on overall population and outdoor worker health Future warming will lead to large adverse human health and well-being impacts, even if we assume there is no further tropical deforestation - an unrealistically conservative baseline given current trends (25). In 2018, ~19.29 million people living in deforested areas across the tropics lost more than 0.5 safe work hours per day, and ~1.4 million people in recently deforested areas lost more than 2 safe working hours per day (Table S1). Global climate change of +1, +1.5, and +2°C beyond 2018 temperatures will increase the number of people in recently deforested locations exposed to 0.5 unsafe work hours per day by 5, 6.5, and 7.7 million people, respectively, even without accounting for future population growth. Climate change similarly increases the number of people exposed to more than 2 lost safe work hours by 2.8, 5.6, and 9.6 million people in these areas (Table S1). Additional global climate change impacts outdoor workers in recently deforested areas as well, with the number of people exposed to >0.5 lost safe work hours per day increasing by 0.35, 0.47, and 0.55 million people under warming of +1, +1.5,

5

and +2°C, respectively (Table S2), for the subset of countries where we have data on outdoor workers. Figure 4 shows results for the fraction of the population living in recently deforested regions that will experience >2 unsafe work hours lost with and additional +1, +1.5, and +2°C of future global warming. Global warming will exacerbate the impacts of deforestation-induced local warming, particularly in tropical Central and South America, eastern Africa, and much of tropical Asia. Safe work hours lost in recently deforested areas increase by about half an hour per degree of global warming (Table S3). Extensive deforestation exacerbates impacts: The case of Amazonian deforestation in Mato Grosso and Pará states We examine Mato Grosso and Pará states in Brazil to illustrate how deforestation-induced temperature effects are pronounced in areas that have experienced extensive tropical deforestation (Figure 5; brief contextual background in SI). Both states are at Brazil’s forest frontier (26), and together they accounted for 63% of all deforestation in the Amazon from 2008-2019, with an average of over 5,000 km2 of forest cleared per year (27). Our results indicate that in these states, 335,914 km2 (20.6% of land area) was deforested before 2003 and 166,890 km2 (10.2% of land area) was deforested between 2003 and 2018, whereas 1,123,850 km2 kept forest cover (68.8% of the area) and 7,021 km2 gained forest cover (0.4% of the area). Temperature increases over 2°C in these two states were found in 57.9% of recently deforested locations compared to just 7.6% of forested locations (Table S4). In total, 215,460 km2 of the Amazon forest region of Mato Grosso and Pará states - an area approximately the size of France – experienced increases of >0.5 hours per day of lost safe work since 2003. In our study period, 46.6% of the deforested area experienced >0.5 hours per day of lost safe work compared with only 4.1% of the area that maintained forests. Approximately 37.8% (~2.64 million people in 2015) of the two states combined population are in areas that lost >0.5 hours per day of safe work between 2003 and 2018. Outdoor workers (here, defined as workers in the agricultural, forestry, and fishery sector) in particular are at risk of excess heat exposure, and across Mato Grosso and Pará state, these workers accounted for 20.5% of the population in these states (~1.41 million people). There are 3,632 km2 of deforested area where more than two hours of safe work hours have been lost since 2003, impacting 2,922 people. With future global warming of +1, +1.5, and +2°C under the assumption of no further deforestation, deforested areas where more than two hours of safe work hours are lost per day will increase to 37,112, 56,948 and 82,941 km2, respectively. Assuming no further population growth, this will lead to 144,010, 213,914, and 288,239 people losing more than two hours per day of safe work hours in deforested areas. Compared to our analysis of Brazil overall, the warming impacts from deforestation in Mato Grosso and Pará state are proportionally larger. For example, when considering regions where more than two hours per day were lost in forested and deforested areas since 2003, Mato Grosso and Pará state had 2.3 times more relative land area and 3.9 times more relative people impacted than Brazil overall.

6

Discussion Heat exposure is already impacting work productivity across the tropics, with climate change expected to reduce ‘workability’ for nearly 1 billion people by the end of the century (e.g., refs. 14,30,31,32,33). Yet research so far has overlooked how deforestation and the loss of cooling services will impact some of the most vulnerable and least resilient populations in the world - rural communities in low latitude countries. This study provides estimates of population impacts of combined heat exposure from recent deforestation and climate change. We show that just 15 years of deforestation (2003-2018) is associated with local warming in many locations equivalent to or surpassing a century of unabated global warming (Figure 1b, ref. 5). We find this warming is associated with widespread expansion of conditions less conducive for safe heavy physical outdoor labor for large portions of the population in the tropics (Figure 1c; Figure 2; Figure 3), although in some locations light labor could be scheduled during the hottest part of the day, which we have not examined here. Warming from deforestation is the most pronounced in tropical South America, where large contiguous patches of deforestation are driven by the expansion of industrial agriculture. By contrast, in Africa temperature changes associated with deforestation are less extreme, which is consistent with a slower expansion of industrial agriculture and a majority of smaller farm operations in Africa (e.g., ref. 5,25). Tropical deforestation is hastening the arrival of heat effects on workability for some of the most vulnerable populations in the world in countries that are expected to bear the brunt of adverse climate change impacts. Our results indicate that over 1.4 million people already experience at least two hours of lost safe work hours per day in recently deforested areas in low-latitude countries. The majority of these people live in low-latitude Asian countries (Table S1), where there is relatively high population density in recently deforested tropical forest biome. With an additional 2°C of global warming, the number of people exposed to two hours of lost safe work hours per day in recently deforested will increase by over 9.6 million, without taking into account likely increases in local populations or further deforestation. As with other research that evaluates population exposure of environmental changes or threats (e.g., Ravinshankara, 2020; Dunne et al., 2013; Liu et al., 2017; Xie et al., 2020; Xu et al., 2020), our approach provides a reasonable estimate for the number of people exposed across the tropics to unsafe thermal environments. We hope these estimates motivate further research to provide more granular estimates. Large portions of the population across the tropics and subtropics currently work outdoors throughout the year (Table S4) and are therefore likely to be impacted by the temperature increases presented here. In American, African, and Asian low-latitude countries, ~15.5%, ~55.3%, and ~56.1%, respectively, of the population work in the agriculture, fisheries, and forestry sectors (Integrated Public Use Microdata Series, ref. 34), although these numbers may underestimate outdoor worker exposure because outdoor work is not limited to these sectors and representative data on outdoor workers for all countries in our sample are limited. Those engaged in informal labor, such as water and fuelwood collection for the household, may also work in hot environments. In addition to heat exposure impacts on work productivity (e.g., ref. 35), heat exposure can cause numerous downstream health impacts. Work in deforested versus forested settings in Indonesia is associated with heat strain (19), and individuals working under heat stress are more likely to experience occupational heat strain and kidney disease or acute kidney injury (13). Specifically, heat stress and recurrent dehydration may contribute to a global epidemic of downstream health impacts in the agriculture industry, such as chronic kidney disease (36) and acute kidney injury (37). Additionally, thermal overload has been suggested as a triggering factor in the death of sugarcane workers (38,39). Globally, high ambient temperatures also increase mortality (40), with large increases in heat-attributable excess mortality projected in tropical countries due to global climate change (11,41). Heat exposure can also interact with other factors to amplify adverse health outcomes, such as traumatic injuries among outdoor working populations (42) and increases in heat-related illnesses (43). Taken together, these findings suggest that deforestation

7

may exacerbate the effect of climate change on the well-being of individuals and families that rely on outdoor work for their livelihood.

Although heat exposure impacts populations around the globe (e.g., ref. 44), subnational analyses of deforestation-associated local warming highlight geographies that are of particular concern - namely, those that have experienced significant deforestation in the past few decades. The magnitude of deforestation-induced warming reported here (Figure 1b) and elsewhere (5) has implications for economic impacts of labor lost (e.g., ref. 45) and raises questions about long-term adaptation to global warming for these hard-hit regions. Beyond the possible downstream health impacts outlined above, local warming exacerbated by deforestation may have implications for migration (46,47), economic production (48), and lower human capital (49).

Tropical deforestation remains an issue of global concern for global climate change and biodiversity conservation (50,51,52,53), and a growing body of work has highlighted how deforestation is driving large-scale warming (5,15). Our results indicate warming from tropical deforestation already impacts safe work hours across the tropics, and these impacts will be exacerbated with future warming. There is an urgent need for action to bolster the climate resilience of rural populations in tropical low-income countries where populations are often identified as contributing the least to climate change, but are expected to bear a disproportionate burden of its negative effects (54,55). Activities that drive deforestation, such as the expansion of agriculture or logging, are relatively predictable, and can be mitigated by proactive land use planning that takes into account the cooling services provided by trees. Recent studies indicate that tropical forest cover gains are associated with land surface temperature decreases (Prevedello et al., 2019; Alkama and Cescatti, 2016). Therefore, forest protection, reforestation, agroforestry on existing agricultural lands, and agricultural intensification with zero deforestation commitments may all be viable strategies that maintain or add trees that can provide cooling services that benefit local populations. Indeed, our results indicate that tree cover gain is associated with temperature decreases and increases in safe work hours (Figure 1b, 1c). These activities also serve as natural climate solutions (56) - that is, actions that can address climate mitigation and adaptation needs. Policy efforts, such as the United Nations’ Decade of Ecosystem Restoration, may serve as a pathway to accelerating this process.

8

Materials and Methods Our primary analytic framework (Figure S1 in SI) combines spatially explicit datasets on forest cover, temperatures, and population characteristics and counts, as well as established relationships between temperatures and safe work hours. Below, we outline each component of our analytic framework, and the analytic methods used to estimate the impact of increased temperatures on safe work hours. All data are interpolated to a common 1km horizontal spatial resolution (the unit of analysis) to facilitate comparison. The following methods are employed at all spatial scales (e.g., global, national, state-level, and local analyses) using data at this common 1km spatial scale. Countries in analysis. Our study examines the local warming impacts from low-latitude (here referred to as ‘tropical’) deforestation (or lack thereof) and future global climate change, and focuses on countries that are within 30° of the equator and have tropical forest biome present within their borders with greater than 75% forest cover present within their borders in the year 2000. As a result, we have 94 countries in our analytic sample, with 34 countries in North, Central, and South America (heretofore the Americas), 31 countries in Africa, and 27 countries in Southeast Asia (heretofore referred to as Asia). Table S6 in SI presents the list of countries. These countries represent 3.98 billion people, of which 7.59% (~300 million) live in tropical forest biome (ref. 20). Forest cover, loss, and gain. We use tree cover data from ref. (21) to examine regions that maintained, lost, and gained forest cover in the tropics and subtropics (30°S-30°N) within the tropical and subtropical forest biome defined in ref. (20). The original forest loss data from ref. (21) have been updated annually since 2013, with data up to and including 2019 in the v1.7 dataset used here. As in ref. (21), here we define a location as forested if a location was >75% forested in the year 2000 (first year of forest data coverage) and contains tropical forest biome (20). The forested location was then reclassified as ‘maintained’ when any forested location did not experience any forest loss over the time period of analysis (2003-2018); ‘loss’ when it lost cover and showed no forest gain (2003-2018); and ‘gain’ when it experienced forest gain and no loss. The forest gain dataset (ref. 21) does not include the exact year of forest cover gain; this dataset only contains binary information about tree gain at each location between the years 2000 and 2012. Therefore, we cannot quantitatively assess the decrease in temperatures and heat stress due to forest gain over this period (Figure 1b, 1c). The effects of the geographic scale of deforestation have not been accounted for here, and may explain much of the variability in temperature changes associated with deforestation (e.g. ref. 5). Forest cover data are provided at a 30m spatial resolution. We use bilinear weighting to regrid deforestation data to the 1km resolution MODIS data so the dataset could be compared at the same grid resolution (‘interp.surface.grid’ in the R ‘fields’ package V. 10.3). Population data and impacts. Our analyses estimate changes to local temperatures from deforestation and the subsequent effects on heat exposure and lost safe work hours for heavy outdoor labor. This methodology follows on past large-scale studies of broadscale effects of environmental conditions (e.g., PM2.5, heat exposure), by estimating the number of people in the areas experiencing notable environmental changes (e.g., Ravinshankara, 2020; Dunne et al., 2013; Liu et al., 2017; Xie et al., 2020; Xu et al., 2020). Table S6 in the SI presents the countries included in the global safe work hours loss analysis and the analysis on outdoor workers. We estimate changes in heat exposure and lost safe work hours between 2003 and 2018 for the general population and for outdoor workers using the fourth version of the Gridded Population of the World (GPW) and IPUMS Terra dataset (22). The GPW provides spatially explicit population data at approximately 1km2 resolution based on population counts between 2005 and 2014 (for details see http://sedac.ciesin.columbia.edu/data/collection/gpw-v4), distributing populations evenly across administrative unit. The main conclusions (e.g., overall population exposure to heat in deforested locations relative to locations that maintained forest cover) remain robust if in place of GPW data we use LandScan data (https://landscan.ornl.gov/) that locates populations in 1km2 pixels instead of spreading these populations across administrative zones (Figure S6). We use the 2015 GPW population data along with the IPUMS Terra data (24; MPC 2018) to estimate impacts of deforestation-associated temperature changes on workers in the agriculture, fishing,

9

and forestry sector that are part of the labor force (referred to as ‘outdoor workers’ in text). IPUMS Terra extracts information from census and survey microdata on individuals and households to provide spatially explicit human population characteristics, such as employment in specific sectors. Industrial classifications are categorized into twelve occupational groups, and are adapted to roughly meet the groupings outlined in the International Standard Industrial Classifications system. Our primary variable of interest is the proportion of the population in ages 16-65 employed in agriculture, fishing, or forestry sectors in a given geographic unit (e.g., county). We apply this percentage to the corresponding geographic unit from the GPW population data that are between 16-65 years of age to get the number of people in any given geographic unit that is working in the agriculture, fishing, or forestry sector. It is possible that this process contains some negligible errors that resulted from administrative boundary mismatch and data conversion from vector to raster. Note that estimates of outdoor worker numbers represent a conservative estimate of the number of people working outside. Our estimates assume people’s daytime locations on average correspond to the 1km2 pixels to which global spatial population datasets assign them. Any assumptions about travel distance that extend beyond this 1km2 area would not be based on observations at the scale necessary. Outdoor worker impact estimates are limited to countries that have data on industrial classifications, which are available for 41 countries (21 from the Americas, 13 from Africa, and 7 from Asia). These countries represent 44% of our total sample, where we maintain 60%, 37%, and 30% of our sample from the Americas, Africa, and Asia, respectively. The GPWv4 population data are several years old (2015), and outdoor worker statistics (IPUMS Terra, ref. 24) do not cover all countries containing tropical forest biome in the tropics (Table S6). However, the main conclusions of our results do not change if we use the LandScan population data in place of the GPWv4 data (Figure S6). Diurnal cycles of temperature and humidity. The National Oceanic and Atmospheric Administration (NOAA) uses heat index, which includes temperature and relative humidity, as a criterion for issuing safety warnings and predicting health impacts associated with heatwaves. Although the MODIS satellite dataset (ref. 57) provides twice daily values for temperature, higher temporal resolution temperature and humidity data are needed to estimate work hours lost. Following the methodology of ref. (16), we use the fifth generation of the European Centre for Medium-Range Weather Forecast atmospheric reanalysis data product (ERA5, ref 58) to obtain the diurnal (24-hour) cycle of temperature and relative humidity over tropical land areas in the years 2003 and 2018. For each grid point over land (30°S-30°N, 120°W-170°E), we calculate the average hourly temperature and relative humidity in 2003 and 2018 on a monthly basis. We use monthly ERA5 data to find the three hottest months for each grid point over tropical land areas (1989-2018). Although we use 2-m air temperature to define the hottest months of the year, if we use heat index to define the hottest months, the results are nearly identical (Figure S7). We then use the average monthly diurnal cycles from these three hottest months to estimate temperature change and heat index changes from deforestation. We choose the 3 hottest months of the year because tropical and subtropical forest biome extends into regions characterized by more pronounced seasonality (Figure 1a). An analysis of ERA5 hourly data shows that inclusion of 12-month mean temperatures at these locations would mask temperature impacts by averaging cool and warm seasons (Figure S8). We use any MODIS 8-day composite observations that overlap with the 3 hottest months of the year from the ERA5 climatology to define the maximum (MODIS Aqua) and minimum (MODIS Terra) values of the diurnal cycle of temperature. Our main results are nearly identical if we analyze more months of the year (Figure S2 in SI), but we show results for the three hottest months because although outdoor work occurs throughout the year in low-latitude countries (Table S5 in SI), agricultural workers must work outside in these months. We use the local specific humidity from the ERA5 data combined with the diurnal cycle of temperature and the assumption that the amount of water vapor in the atmosphere was constant, and calculate relative humidity accordingly (16). With the shape of the diurnal cycle from ERA5, specific humidity from ERA5, and the two values for daytime and nighttime temperatures obtained from MODIS data, we generate hourly estimates of heat indices and estimate the total time per day spent above the heat stress threshold limit for heavy physical work (see Heat index distributions below).

10

Although 2-m air temperature is often used to calculate heat index, here we rely on land surface temperature from satellite data because 2-m air temperature is not available at the necessary spatial resolution to conduct this analysis. Although some studies (e.g., ref. 15) have used a complex modeling approach to estimate 2-m air temperature from land surface temperature, here we focus on differences in land surface temperature before and after deforestation, which should help eliminate much of the 2-m vs surface temperature differences. Recent studies have found that canopy air temperatures in tropical forests are between 2 and 5˚C warmer than the two-meter air temperatures beneath the canopy (e.g., ref. 59), and an analysis of tower data taken over a tropical pasture during the GOAMAZON campaign (not shown) shows land surface temperatures that are measured by the MODIS satellite are generally 2-6˚C warmer than the local two-meter air temperature. Because both of our available temperature measurements are roughly equally biased towards warmer values that are strongly correlated with 2-m air temperature, we do not expect our estimates of two-meter air temperature differences between forested and deforested regions to be meaningfully biased. The short satellite record provides a limited (~18 year) perspective on local climate change from deforestation, but we are forced to analyze year to year variability, which introduces weather-scale noise into our analysis (Figure 1b). Nonetheless, the main findings from these deforestation-associated temperature change results remain robust if different years are analyzed (e.g., 2004 vs 2017; Figure S3 in SI), indicating that temperature and associated work hours lost results are not due to year to year, natural climate variability. Heat index distributions. Following the methodology of ref. (16), diurnal heat index cycles were estimated from diurnal cycles of temperature and relative humidity (see Diurnal cycles of temperature and humidity above). Heat indices were estimated using a refinement of Rothfusz’s multiple linear regression analysis of Steadman’s complex, multi-parameter equations (See https://www.wpc.ncep.noaa.gov/html/heatindex_equation.shtml for the equations used). The heat index value for the ACGIH Threshold Limit Value (TLV) for heat stress, based on existing literature for heavy physical work in agriculture or construction is 85oF (ref. 10). In the text, we report ‘work hours lost’ as the total time per day spent above this 85oF heat index threshold. Our analysis focuses on using heat index values rather than Wet Bulb Globe Temperature (WBGT, a measure of heat stress) values as using WBGT is impractical in our study contexts - calculation of WBGT requires estimates of temperature, humidity, wind speed, sun angle, and cloud cover. Ref. (60) and more recently ref. (61) evaluated the use of HI values compared to WGBT, indicating that HI provides a reasonable alternative to WBGT when WBGT is not practical or available. As in ref. (16), we assume workers are acclimatized, able to rest in the shade, utilize regular single-layer work clothes, and are conducting a 415 W metabolic rate work, based on literature for heavy physical work in agriculture or construction (10,62). For work hours lost estimates, we used the same underlying assumptions (e.g., work intensity, clothing) for all industries. Although our results do not provide estimates of work productivity by specific industries, occupations, or tasks (which drive absolute heat stress levels), the relative changes in hours lost due to heat exposure from deforestation and climate change in different populations and geographical areas reported here provide important information about populations at highest risk. Future temperature changes. Following the methodology of ref. (16), we used output from 31 CMIP5 models to estimate local surface temperature and specific humidity changes per degree of global warming (‘Pattern scaling’ method, ref. 63) under a ‘high’ climate change scenario (Representative Concentration Pathway 8.5). The method allows us to estimate local temperature and humidity changes under an additional 1℃, 1.5℃, and 2℃ of warming to derive future estimates of heat indices and hours lost. The RCP8.5 experiment can include land use changes, so local temperature and humidity changes from RCP8.5 were compared to changes from 1%CO2 experiments, and multi-model mean results were found to be nearly identical (Figure S4 in SI). We have tested the sensitivity of our results using the CMIP5 skin surface temperature (‘ts’) variable and the 2-m air temperature variable (‘tas’) and find no significant changes in our results (not shown). Regional pattern scaling results are shown in Figure S5. We limited our analysis to the 31 models that provided surface temperature and specific humidity output: ACCESS1-0, ACCESS1-3, bcc-csm1-1-m, bcc-csm1-1, BNU-ESM, CCSM4, CESM1-BGC, CESM1-CAM5, CMCC-CESM, CMCC-CMS, CMCC-CM, CNRM-CM5, CSIRO-Mk3-6-0,

11

CanESM2, EC-EARTH, FGOALS-g2, FGOALS-s2, FIO-ESM, GFDL-CM3, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-H, GISS-E2-R, HadGEM2-AO, HadGEM2-CC, HadGEM2-ES, inmcm4, IPSL-CM5A-LR, IPSL-CM5A-MR, IPSL-CM5B-LR, MIROC-ESM-CHEM, MIROC-ESM, MIROC5, MPI-ESM-LR, MPI-ESM-MR, MRI-CGCM3, MRI-ESM1, NorESM1-ME, NorESM1-M.

12

References 1. E. S. Brondizio, J. Settele, S. Díaz, H. T. Ngo, Global assessment report on biodiversity

and ecosystem services of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services (2019).

2. P. G. Curtis, C. M. Slay, N. L. Harris, A. Tyukavina, M. C. Hansen, Classifying drivers of global forest loss. 361, 1108–1111 (2018).

3. D. Ellison, et al., Trees, forests and water: Cool insights for a hot world. 43, 51–61 (2017). 4. C. E. Scott, et al., Impact on short-lived climate forcers increases projected warming due

to deforestation. 9, 1–9 (2018). 5. L. V. Zeppetello, et al., Large scale tropical deforestation drives extreme warming (2020). 6. A. S. Cohn, et al., Forest loss in Brazil increases maximum temperatures within 50

km. 14, 084047 (2019). 7. C. Xu, T. A. Kohler, T. M. Lenton, J.-C. Svenning, M. Scheffer, Future of the human climate

niche (2020). 8. E. D. Coffel, R. M. Horton, A. de Sherbinin, Temperature and humidity based projections

of a rapid rise in global heat stress exposure during the 21st century. 13, 014001 (2017). 9. M. McKinnon, et al., Climate Change and Labour: Impacts of Heat in the

Workplace. 29 (2016). 10. (ACGIH) American Conference of Governmental Industrial Hygienists, Heat stress and

strain: TLV® Physical Agents (2009). 11. C. Mora, et al., Global risk of deadly heat. 7, 501–506 (2017). 12. J. T. Spector, P. E. Sheffield, Re-evaluating occupational heat stress in a changing

climate. 58, 936–942 (2014). 13. A. D. Flouris, et al., Workers’ health and productivity under occupational heat strain: a

systematic review and meta-analysis. 2, e521–e531 (2018). 14. T. Kjellstrom, et al., Climate change and occupational heat problems. 51, 1–2 (2013). 15. R. Alkama, A. Cescatti, Biophysical climate impacts of recent changes in global forest

cover. 351, 600–604 (2016). 16. N. Wolff, et al., The effect of deforestation and climate change on all-cause mortality and

unsafe work conditions due to heat exposure in Indonesia (submitted). 17. N. H. Wolff, Y. J. Masuda, E. Meijaard, J. A. Wells, E. T. Game, Impacts of tropical

deforestation on local temperature and human well-being perceptions. 52, 181–189 (2018). 18. Y. J. Masuda, et al., How are healthy, working populations affected by increasing

temperatures in the tropics? Implications for climate change adaptation policies. 56, 29–40 (2019).

19. M. K. Suter, et al., Association between work in deforested, compared to forested, areas and human heat strain: an experimental study in a rural tropical environment. 14, 084012 (2019).

20. E. Dinerstein, et al., An ecoregion-based approach to protecting half the terrestrial realm. 67, 534–545 (2017).

21. M. C. Hansen, et al., High-resolution global maps of 21st-century forest cover change. 342, 850–853 (2013).

22. C. U. [CIESIN] Center for International Earth Science Information Network, I. F. P. R. Institute, W. R. Institute, Gridded Population of the World, Version 4 (GPWv4): Administrative Unit Center Points with Population Estimates (2016).

23. X. Liu, et al., High-spatiotemporal-resolution mapping of global urban change from 1985 to 2015. 1–7 (2020).

24. S. Ruggles, T. A. Kugler, C. A. Fitch, D. C. Van Riper, Terra Populus: Integrated data on population and environment in (IEEE, 2015), pp. 222–231.

25. X.-P. Song, et al., Global land change from 1982 to 2016. 560, 639–643 (2018). 26. G. M. Thaler, C. Viana, F. Toni, From frontier governance to governance frontier: The

political geography of Brazil’s Amazon transition. 114, 59–72 (2019).

13

27. L. F. FG Assis, et al., TerraBrasilis: A Spatial Data Analytics Infrastructure for Large-Scale Thematic Mapping†. 8 (2019).

30. J. P. Dunne, R. J. Stouffer, J. G. John, Reductions in labour capacity from heat stress under climate warming. 3, 563–566 (2013).

31. T. Kjellstrom, R. S. Kovats, S. J. Lloyd, T. Holt, R. S. Tol, The direct impact of climate change on regional labor productivity. 64, 217–227 (2009).

32. Y. Chavaillaz, et al., Exposure to excessive heat and impacts on labour productivity linked to cumulative CO 2 emissions. 9, 1–11 (2019).

33. D. Li, J. Yuan, R. B. Kopp, Escalating global exposure to compound heat-humidity extremes with warming (2020).

34. S. Flood, M. King, R. Rodgers, S. Ruggles, J. R. Warren, Integrated Public Use Microdata Series, IPUMS, (2018).

35. S. Sahu, M. Sett, T. Kjellstrom, Heat exposure, cardiovascular stress and work productivity in rice harvesters in India: implications for a climate change future (2013).

36. V. M. Weaver, J. J. Fadrowski, B. G. Jaar, Global dimensions of chronic kidney disease of unknown etiology (CKDu): a modern era environmental and/or occupational nephropathy? 16, 145 (2015).

37. U. P. Santos, D. M. T. Zanetta, M. Terra-Filho, E. A. Burdmann, Burnt sugarcane harvesting is associated with acute renal dysfunction. 87, 792–799 (2015).

38. B. F. A. OLIVEIRA, et al., Human Heat stress risk prediction in the Brazilian semiarid Region based on the Wet-Bulb Globe Temperature. 91 (2019).

39. D. P. Bitencourt, Á. C. Ruas, P. A. Maia, Analysis of the contribution of meteorological variables to thermal stress associated with death among sugarcane cutters 28, 65–74 (2012).

40. A. Gasparrini, et al., Mortality risk attributable to high and low ambient temperature: a multicountry observational study. 386, 369–375 (2015).

41. J. Y. Lee, et al., Predicted temperature-increase-induced global health burden and its regional variability. 131, 105027 (2019).

42. J. T. Spector, et al., A case-crossover study of heat exposure and injury risk in outdoor agricultural workers. 11, e0164498 (2016).

43. D. L. Riquinho, E. A. Hennington, Health, environment and working conditions in tobacco cultivation: a review of the literature. 17, 1587–1600 (2012).

44. A. Gasparrini, et al., Mortality risk attribuTable So high and low ambient temperature: a multicountry observational study. 386, 369–375 (2015).

45. A. Orlov, J. Sillmann, K. Aunan, T. Kjellstrom, A. Aaheim, Economic costs of heat-induced reductions in worker productivity due to global warming. 63, 102087 (2020).

46. S. M. Hsiang, A. H. Sobel, Potentially extreme population displacement and concentration in the tropics under non-extreme warming. 6, 25697 (2016).

47. V. Mueller, C. Gray, K. Kosec, Heat stress increases long-term human migration in rural Pakistan. 4, 182–185 (2014).

48. T. A. Carleton, S. M. Hsiang, Social and economic impacts of climate. 353, aad9837 (2016).

49. T. Garg, M. Jagnani, V. Taraz, Temperature and human capital in India (2018). 50. P. R. Shukla, et al., IPCC, 2019: Climate Change and Land: an IPCC special report on

climate change, desertification, land degradation, sustainable land management, food security, and greenhouse gas fluxes in terrestrial ecosystems.

51. R. A. Mittermeier, N. Myers, J. B. Thomsen, G. A. Da Fonseca, S. Olivieri, Biodiversity hotspots and major tropical wilderness areas: approaches to setting conservation priorities. 12, 516–520 (1998).

52. B. T. Smith, et al., The drivers of tropical speciation. 515, 406–409 (2014). 53. M. Gloor, et al., The carbon balance of South America: a review of the status, decadal

trends and main determinants. 9, 5407–5430 (2012).

14

54. L. J. Harrington, et al., Poorest countries experience earlier anthropogenic emergence of daily temperature extremes. 11, 055007 (2016).

55. A. D. King, L. J. Harrington, The inequality of climate change from 1.5 to 2 C of global warming. 45, 5030–5033 (2018).

56. B. W. Griscom, et al., Natural climate solutions. 114, 11645–11650 (2017). 57. NASA Goddard Space Flight Center, Ocean Ecology Laboratory, Ocean Biology

Processing Group. Moderate-resolution Imaging Spectroradiometer (MODIS) Terra Data; NASA OB.DAAC, Greenbelt, MD, USA.

58. Copernicus Climate Change Service (C3S) (2017): ERA5: Fifth generation of ECMWF atmospheric reanalyses of the global climate . Copernicus Climate Change Service Climate Data Store (CDS), date of access. https://cds.climate.copernicus.eu/cdsapp#!/home

59. T. Kumagai, et al., Vertical profiles of environmental factors within tropical rainforest, Lambir Hills National Park, Sarawak, Malaysia. 6, 257–264 (2001).

60. A. W. Tustin, et al., Evaluation of occupational exposure limits for heat stress in outdoor workers—United States, 2011–2016. 67, 733 (2018).

61. X. P. Garzón-Villalba, C. D. Ashley, T. E. Bernard, Benchmarking Heat Index as an occupational exposure limit for heat stress. 16, 557–563 (2019).

62. T. Kjellstrom, C. Freyberg, B. Lemke, M. Otto, D. Briggs, Estimating population heat exposure and impacts on working people in conjunction with climate change. 62, 291–306 (2018).

63. B. D. Santer, T. M. Wigley, M. E. Schlesinger, J. F. Mitchell, Developing climate scenarios from equilibrium GCM results (1990).

Additional References Cited in Revision

Ellison, D. et al. Trees, forests and water: Cool insights for a hot world. Global Environ. Change 43, 51-61 (2017).

Liu, Z. et al. Global and regional changes in exposure to extreme heat and the relative contributions of climate and population change. Scientific Reports 7, 43909 (2017).

Masuda, Y. et al. Heat exposure from tropical deforestation decreases cognitive performance of rural workers: an experimental study. Environmental Research Letters (2020).

Masuda, Y. J. et al. How are healthy, working populations affected by increasing temperatures in the tropics? Implications for climate change adaptation policies. Global Environ. Change 56, 29-40 (2019).

McAlpine, C. A. et al. Forest loss and Borneo’s climate. Environmental Research Letters 13, 044009 (2018).

Prevedello, J. A., Winck, G. R., Weber, M. M., Nichols, E. & Sinervo, B. Impacts of forestation and deforestation on local temperature across the globe. PloS one 14, e0213368 (2019).

Ravishankara, A. R., David, L. M., Pierce, J. R. & Venkataraman, C. Outdoor air pollution in India is not only an urban problem. Proceedings of the National Academy of Sciences (2020).

Xie, Y. et al. Health and economic benefit of China’s greenhouse gas mitigation by 2050. Environmental Research Letters 15, 104042 (2020).

15

Acknowledgments LAP thanks the WRF Postdoctoral Fellowship and NASA for funding support. This work contains modified Copernicus Climate Change Service Information. The forest cover could not have been analyzed without NASA and the USGS partnership that resulted in the Landsat mission. Support for this work was provided by a pilot research grant from the University of Washington Population Health Initiative. Author Contributions LAP, NHW, LVZ, DSB, TK, YJM, and JTS designed the research. JJ, LAP, TK, NHW and LVZ conducted the data analysis, and LAP, JJ, NHW, LVZ, DSB, TK, YJM, and JTS interpreted data analysis outputs. LAP, JJ, NHW, LVZ, DSB, TK, YJM, and JTS wrote the paper. Competing Interest Statement The authors declare no competing interests.

16

Figures

Figure 1. Map (A) show tropical and subtropical forest biome and distributions (B, C) show temperature changes and safe work hours lost differences 2003-2018. Map (A) illustrates regions with forest cover in the year 2000 (grey, ref. 21), regions deforested between 2003 and 2018 (red, ref. 21), and locations with potential tropical and subtropical forest biome that were deforested before the start of the study period (green, ref. 20). Distributions (B, C) show differences in temperature (B) and differences in hours lost (C) in the three main tropical forest regions between 2018 and 2003. Grey curves show PDFs of areas that maintained forest cover (‘kept forest’), red curves show locations that lost forest cover (‘lost forest’), and blue curves show locations that gained forest cover (‘gained forest’). PDFs showing changes in hours lost per day show distributions from locations with greater than 0.25 hours lost per day in 2003 and 2018 (PDFs with all locations shown in Figure S9). Americas includes 34 countries, Africa includes 32, and SE Asia includes 27 countries (full list of countries in Table S6 in SI).

17

Figure 2. Percent of land area (A) and percent of population (B) that experienced conditions creating exposure to increases in loss of safe work hours between 2003 and 2018 in forested and deforested areas in the Americas, Africa, and Asia.

18

Figure 3. Population living in deforested areas (in thousands) and percent of population living in deforested areas where more than 0.5 hours of safe work hours were lost between 2003 and 2018.

19

Figure 4. Percent of population living in deforested areas exposed to > 2 hours of safe work hours lost in deforested areas in 2018 and with an additional 1℃, 1.5℃, and 2℃ of global warming.

20

Figure 5. Forest cover change, temperature change, lost safe work hours change, and distribution of lost safe work hours for locations that maintained forest, lost forest, and gained forest in Mato Grosso and Para states in Brazil. Differences shown between 2003 and 2018. PDFs showing changes in hours lost per day in (e) show distributions from locations with greater than 0.25 hours lost per day in 2003 and 2018 (PDFs with all locations shown in Figure S10).