A Century of Reforestation Reduced Anthropogenic Warming in the Eastern United States

Restoring and preserving the world's forests are promising natural pathways to mitigate some aspects of climate change. In addition to regulating atmospheric carbon dioxide concentrations, forests modify surface and near‐surface air temperatures through biophysical processes. In the eastern United States (EUS), widespread reforestation during the 20th century coincided with an anomalous lack of warming, raising questions about reforestation's contribution to local cooling and climate mitigation. Using new cross‐scale approaches and multiple independent sources of data, we uncovered links between reforestation and the response of both surface and air temperature in the EUS. Ground‐ and satellite‐based observations showed that EUS forests cool the land surface by 1–2°C annually compared to nearby grasslands and croplands, with the strongest cooling effect during midday in the growing season, when cooling is 2–5°C. Young forests (20–40 years) have the strongest cooling effect on surface temperature. Surface cooling extends to the near‐surface air, with forests reducing midday air temperature by up to 1°C compared to nearby non‐forests. Analyses of historical land cover and air temperature trends showed that the cooling benefits of reforestation extend across the landscape. Locations surrounded by reforestation were up to 1°C cooler than neighboring locations that did not undergo land cover change, and areas dominated by regrowing forests were associated with cooling temperature trends in much of the EUS. Our work indicates reforestation contributed to the historically slow pace of warming in the EUS, underscoring reforestation's potential as a local climate adaptation strategy in temperate regions.


Introduction
Drastic reductions in anthropogenic greenhouse gas emissions are necessary to address climate change.Naturebased Climate Solutions (NbCS), such as reforestation, have the potential to provide additional mitigation through atmospheric CO 2 removal (Nolan et al., 2021;Novick, Metzger, et al., 2022;Novick, Williams, et al., 2022;Seddon et al., 2020), but will only be effective if they are accompanied by economy-wide decarbonization.Changes to land cover and management, which are central to the implementation of NbCS, can alter local temperature through changes to the surface energy balance (Anderson et al., 2011).If these biophysical impacts are beneficial, then some NbCS could serve as a tool for local adaptation in addition to global-scale climate mitigation.
Reforestation-the NbCS with the highest CO 2 mitigation potential (Griscom et al., 2017)-can increase or decrease local surface temperature (T s ) depending on the balance of competing mechanisms.In the tropics, forests evaporate substantially more water than grasslands, which promotes cooling by using energy that would otherwise heat the surface (Anderson et al., 2011;Williams et al., 2021).Conversely, reforestation in boreal climates tends to warm the surface due to both reductions in albedo and the influences of surface roughness on nighttime temperatures (Lee et al., 2011).In the temperate zone, surface cooling from increased evaporative and sensible heat fluxes usually outweighs albedo-driven warming, such that temperate forests have lower T s compared to nonforested ecosystems (Anderson et al., 2011;Bright et al., 2017;Burakowski et al., 2018;Windisch et al., 2021;Zhang et al., 2020).
Although T s is relevant to many ecological processes (Farella et al., 2022), the near-surface air temperature (T a ) is an equally important target for climate adaptation (Novick & Katul, 2020;Winckler et al., 2019) because changes in T a can have far-reaching effects, as biophysical impacts on air temperature can be advected across the landscape (Winckler et al., 2019).While land-cover change affects T s and T a differently (Baldocchi & Ma, 2013;Helbig et al., 2021;Novick & Katul, 2020;Winckler et al., 2019), quantifying the impacts of land-cover change on T a has historically been challenging.Near-surface air temperature cannot be sensed remotely, and its variation with height makes it difficult to interpret over differing land cover types (Novick & Katul, 2020;Winckler et al., 2019).Most data-driven studies investigating biophysical impacts on both T s and T a typically compare locations with similar macroclimates but different land cover, which captures the direct, local effects of landcover change on T s and T a (e.g., Baldocchi & Ma, 2013;Bright et al., 2017;Juang et al., 2007;Winckler et al., 2019;Windisch et al., 2021;Zhang et al., 2020).However, this approach can overlook indirect, non-local effects (Baldocchi & Ma, 2013) related to advection or changes in downstream cloud cover due to increased upstream evapotranspiration.This study adopts a novel combination of approaches to evaluate the impacts of reforestation on both T s and T a , exploring both local and larger-scale effects across various spatial and temporal scales.
The Eastern United States (EUS) has undergone extensive reforestation over the last century (Figures 1a and 1b; Ramankutty et al., 2010), providing a unique opportunity to investigate the biophysical impacts of large-scale land cover change.By large scales, we mean scales that are sufficiently large so that the atmospheric boundary layer is in local equilibrium with the land-surface.This equilibrium may be achieved at scales much larger (i.e., order of magnitude) than the daytime boundary layer height (order of 1 km).Additionally, the absence of warming over a large portion of the EUS during this period (Figure 1c; Meehl et al., 2012) raises the question of whether reforestation has dampened the historic pace of warming in the region.To address this question, we employ multiple independent data sources to evaluate both local and non-local effects of reforestation on T s and T a.Our approach involves: (a) comparing locations with similar climates but different land cover using both satellite and in situ observations to determine the local effects of reforestation on T s and T a , (b) exploring gradients in T s across ecosystem boundaries to uncover the potential local extent of such effects, and (c) analyzing historical weather station, air temperature, and land-cover data to identify long-term links between T a and forest cover trends at landscape and regional scales.Through this comprehensive approach, we aim to gain insight into the extent to which EUS reforestation has influenced historical rates of regional warming and the potential of temperate zone reforestation for climate adaptation.

Historic Land Cover and Climate Trends in the Region
Before European settlement, forests occupied most of the land area in the EUS, with an uneven-aged stand structure sustained by selective harvest (Novick, Jo, et al., 2022;Ramankutty et al., 2010).However, from the late 18th to early 20th century, forest cover in the EUS dramatically decreased due to harvesting for timber and clearing for agriculture, resulting in forest losses exceeding 90% in some locations (Carman, 2013;Hall et al., 2002;Houghton & Hackler, 2000).By 1930, widespread land clearing had largely stopped, and forest cover began to increase with the abandonment of marginal agricultural fields and active reforestation efforts (Carman, 2013;Hall et al., 2002;Houghton & Hackler, 2000;Ramankutty et al., 2010, and see Figure 1d).Since 1900, millions of hectares of forest have been added in the northeastern, southeastern, and midwestern US (Figure 1b, and Carman, 2013;Hall et al., 2002;Wear & Greis, 2012), mostly through the conversion of crop and pastureland to deciduous forests, or to pine plantations providing softwood timber in some parts of the southeastern US (Wear & Greis, 2012).Many forests in the region are now 50-100 years old (Figure 1a, Figure S1 in Supporting Information S1), although frequent harvest of pine plantations suppresses stand age in the Southeast.This century of EUS reforestation coincides with an anomalous lack of regional warming, sometimes referred to as a "warming hole" (Figure 1; Mascioli et al., 2017;Meehl et al., 2012;Z. Pan et al., 2004;Partridge et al., 2018;Tosca et al., 2017), especially in the Southeast.While most land areas worldwide warmed during the twentieth century, much of the EUS experienced minor cooling, from 0.2°C to 0.8°C per 50 years (Figure 1c).Proposed explanations for this cooling include internal climate variability (Mascioli et al., 2017;Meehl et al., 2012), anthropogenic aerosols (Tosca et al., 2017), agricultural intensification (Mueller et al., 2016), and increasing precipitation (Z.Pan et al., 2004).However, mechanistic attribution remains elusive (Mascioli et al., 2017;Partridge et al., 2018).The warming hole extends across a large area, predominantly affecting the Eastern and Southeastern US (Figure 1), rather than being confined solely to the Midwest.Previous studies, such as Mueller et al. (2016), have identified local cooling effects on temperature extremes in the Midwest due to agricultural intensification.However, these localized effects on temperature extremes do not fully correspond with the persistent and widespread trends in mean temperatures observed throughout the broader region of the warming hole.Despite the established potential of reforestation to affect local temperature, the impacts of regional reforestation over the past century have not been thoroughly evaluated for their contribution to the "warming hole" in the EUS.

Materials and Methods
To evaluate the influence of EUS reforestation on historical warming rates and assess the potential of temperatezone reforestation as a climate adaptation strategy, we integrated several independent data sources, adopting methods that encompass a range of spatial and temporal scales.Our approaches (outlined in Figure 2) are organized across two axes: biophysical versus historical and local versus regional.In broad terms, our investigation: 1. Analyzes satellite and in situ flux tower observations to quantify the local potential impact of forests, and by extension, reforestation on T s .We accomplish this analysis by comparing co-located ecosystems with similar climates but different land cover (i.e., paired sites), and).We also utilize a method that normalizes the local T s time series to account for changes in macro-scale air temperature, enabling us to isolate the land cover effect.2. Investigates the extent to which changes in T s extend to near-surface air temperatures at local scales, including across ecosystem boundaries.These analyses leverage eddy-covariance flux data collected at meteorological towers, as well as gradients in satellite-derived T s across forest-grassland transitions.These approaches help us estimate the potential magnitude and spatial extent of the biophysical effects of reforestation.3. Integrates historical weather station, climate reanalysis, and land-cover data to trace long-term links between T a and forest cover at both the local and regional scales.This process involves analyzing the long-term trends in T a anomalies and determining correlations between forest age and long-term T a trends.

Overview of the Relevant Temperature Metrics
Land surface (or skin) temperature (T s ) is commonly used to assess the impacts of land-cover change and is observable from both flux towers and remote sensing platforms.In terrestrial ecosystems, T s represents the temperature of the uppermost layer of vegetation, reflecting the outcome of the interactive effects of radiative transfer, leaf energy balance, eco-physiological controls on stomatal opening and closure, and aerodynamics.A summary of the temperature metrics employed in this study can be found in Table 1.
T s is mechanistically connected to near-surface T a through standard boundary layer theory for stratified turbulent flows.In vegetated systems, a thin layer of air called the roughness sublayer lies between the vegetation and the so-called surface layer (i.e., the layer where the log-law for the mean velocity and mean air temperature profiles hold in near-neutral conditions).For comparative (and conventional) purposes, near-surface T a is generally measured (and modeled) at a height of 2 m above the surface, but profiles of T a are influenced by canopy structural The diagram is divided into four quadrants to differentiate local and regional scales and biophysical and historical contexts.Each quadrant outlines the associated methods and their purpose.The methodologies span a broad range of space (local to regional) and time (diurnal to long-term historical trends) scales and enriches biophysical analyses (energy-radiation balance combined with physiological and aerodynamic approaches) with empirical or data-driven approaches.effects within the roughness sublayer.In short-stature ecosystems such as grasslands and croplands, the roughness sublayer is typically 1-2 m thick, such that flux tower measurements of T a are typically made in the surface layer.However, above forests with heights of 10 m or more, the roughness sublayer is much thicker (12-50 m), such that tower T a indicates the mean air temperature within the roughness sublayer, not the surface layer (Novick & Katul, 2020).Thus, comparing T a observations across forested and non-forested ecosystems is fraught with potential bias linked to canopy structural effects and measurement height despite the use of a "reference" height of 2 m.To overcome this limitation, we focus on two proxies for T a that are less impacted by structure effects on the near-surface T a profile and can therefore be inferred from eddy covariance sensible heat flux tower data, as detailed by Novick & Katul, 2020.The first proxy is the aerodynamic temperature (T aero ), which represents the air temperature within the upper reaches of the canopy.The second proxy is the air temperature extrapolated upwards into the lower reach of the surface layer (T extrap ).
At the regional scale evaluating the impacts of land cover on T s requires a strategy to control for background variation in macroclimate.Here, the difference between remotely sensed T s and gridded T a was evaluated, with the latter provided by the Daymet (Thornton et al., 2016) product (T a,Daymet ).Remotely sensed surface temperature is frequently termed "LST" (Land Surface Temperature), but we use T s here for consistency with the other temperature metrics.Daymet interpolates data from meteorological weather stations, which are typically located over short grass surfaces; therefore, it is viewed here as a "reference" T a that is not influenced by variability in land cover.Tower-derived estimates of T s and T a were also normalized by the DAYMET T a product, which allowed us to leverage information from all flux towers in the study region, and not just those from co-located paired sites.

Paired Site Flux Tower Analyses
To mechanistically understand reforestation effects on T s over the diurnal cycle (i.e., Figures 3c and 3d), a paired site approach (e.g., Lee et al., 2011;Zhang et al., 2020) is used, relying on observations from six forest-grassland site pairs in the study region.The site pairs are in Arkansas, North Carolina (3 pairs), Indiana, and New Hampshire (Table S1 in Supporting Information S1), and each paired set is separated by ∼30 km or less.Site descriptions and details on eddy covariance data processing are provided elsewhere (Zhang et al., 2020).Briefly, all data were quality-controlled and gap-filled using community-accepted standards embedded in the REddyProc processing tool (Wutzler et al., 2018).The T s data were inferred from the outgoing longwave radiation using the Stefan-Boltzmann law, with emissivity estimated as an empirical function of albedo following an established approach (Juang et al., 2007).The attribution of changes in T s to relevant mechanisms (i.e., variation in sensible vs. latent heat flux) was accomplished through a Taylor-series expansion of the site-level energy balance equation, as described elsewhere (Zhang et al., 2020).In this study, these prior results are aggregated across the site pairs.
The approach to estimate T aero and T extrap is described elsewhere (Novick & Katul, 2020).Briefly, T aero is determined by first quantifying the mean ratio between T s and the tower-measured T a when the measured sensible heat flux is near zero, implying that T a and T aero should be in equilibrium.The determination of tower-measured T a was performed for each hour of the day at each site separately for the peak of the growing season (June-September) and the dormant season (November-March) and was then used to estimate T aero from the observed T s for every hourly or half-hourly observation period.The T extrap was then calculated assuming the stabilitycorrected logarithmic profiles from Monin-Obukhov Similarity Theory (Monin & Obukhov, 1954) hold.
Those calculations are forced with the estimated T aero and an estimate for the roughness length for heat that varies as a function of measured friction velocity as well as the momentum roughness length that is presumed constant for fully rough conditions (Novick & Katul, 2020).Here, results are shown for the temperature extrapolated into the first 10 m of the surface layer after conceptually "flattening" the ecosystems by replacing them with a rough surface characterized by two roughness heights: one for momentum absorption and one for heat transfer (see Novick & Katul, 2020 for details and code).This manuscript extends the results presented in Novick and Katul (2020) to the full set of paired sites in Table S1 in Supporting Information S1, noting that Novick and Katul only evaluate the results from the Duke Forest sites in North Carolina.
The T a data presented in Figure 3 are mean air temperature data measured from a T a /RH probe (such as the HMP35 or HMP45C, Vaisala) above the canopy, usually at or near the height of the eddy covariance systems (see Table S1 in Supporting Information S1).

Broader Synthesis of Flux Tower Sites
The 58 flux towers used in the broader synthesis (Figures 4a-4d) include the paired tower sites, as well as data sets acquired from the AmeriFlux network (Novick et al., 2018).Specifically included are towers installed and operated by the U.S. National Science Foundation's National Ecological Observatory Network (Metzger   , 2019).The network data were limited to those that adopt a CC-BY-4.0 data use license.We first downloaded all available site-years from the network, excluding wetlands and irrigated croplands within the study region.Some sites and site-years were excluded due to missing information about radiative fluxes that are required to derive or calculate the temperature metrics.The towers retained for this analysis are described in Table S2 in Supporting Information S1.The approaches used to determine T s , T aero , and T extrap for the broader flux synthesis are the same as those described in Section 2.2.1, with the exception that the AmeriFlux data were not gap-filled; rather, data were filtered to exclude those collected under very stable conditions in which turbulence generation is suppressed by buoyancy forces, and to exclude excessively large anomalous observation of sensible heat flux (i.e., >1,000 W/m 2 ).For the purposes of this paper, croplands managed as corn/soy rotations were treated as separate sites for corn and soy years.Estimates of canopy height and measurement height are required to calculate T aero and T extrap .For most towers, they are available from the AmeriFlux Biological, Ancillary Disturbance, and Metadata (BADM) database.In cases where they were not available from the BADM, they were extracted from published studies.In rare cases, a "best" guess was made based on known information about the canopy type or from site photos.
To synthesize information from this much broader set of flux towers, it is necessary to adjust for the influence of background variation in macro-climate, so the land cover impacts can be isolated.Toward this end, we specifically evaluated the difference between the tower-derived temperature metrics and daily maximum T a estimates from the gridded 1 km Daymet product (Thornton et al., 2016), referred to as T s -T a .During the growing season, vegetated surfaces typically have higher surface temperature than air temperature, and thus a positive T s -T a (Mildrexler et al., 2011;Novick & Barnes, 2023).We anticipate a reduced T s -T a difference in forests due to their increased transpiration rates due to non-forests, especially at midday when transpiration peaks.This reduced difference reflects the impact of transpiration on sensible heat fluxes and consequently on T s .To the extent that surface cooling extends to the near-surface air temperature at local scales, we also expect the difference between T s and T aero and T extrap to be lower for forests than grasslands.

Differences in T s -T a by Land Cover
To compare relative surface cooling between different types of land cover on a regional scale, we used a remote sensing approach based on T s retrievals from the Moderate Imaging Resolution Spectroradiometer (MODIS; Wan et al., 2015).These T s retrievals were adjusted for variation in macroclimate using daily maximum T a estimates from Daymet (Thornton et al., 2016).The MODIS Land Surface Temperature/Emissivity Daily Product (MYD11A1v6.1) has a spatial resolution of 1 km.The timing of the Aqua MODIS overpass (∼1:30 p.m. local time) generally corresponds to the timing of daily maximum T s estimated from tower measurements (Figure S2 in Supporting Information S1).Thus, to evaluate spatial variability in the difference between remotely sensed T s , we normalized for macro-scale climate variability by subtracting the daily maximum T a,Daymet from MODIS T s .
The average T s -T a,Daymet for each 1 km pixel in the study area was obtained for each month from 2002 to 2018.The data presented in Figures 3a and 3b are the average monthly T s -T a,Daymet for the full time-series.The land cover type for each 1 km MODIS pixel was determined using the U.S. Geological Survey North American Land-Cover Characteristics 1 km grid-spacing data set, created by the National Center for Earth Resources Observation and Science (EROS) as part of the Global Land Cover Characterization Project (EROS, 2017).This data set, derived from 1 km Advanced Very High-Resolution Radiometer data collected between 1992 and 1993, follows the methodologies described in Loveland et al. (1999).There were relatively few pixels classified as grassland, so we combined croplands and grasslands for the spatial T s -T a, Daymet maps in Figure 3a (see also Figures S3-S5 in Supporting Information S1).

Land Surface Temperature Across Forest/Agriculture Boundaries
The National Land Cover Database (NLCD) land cover data set for 2016 (Dewitz, 2019) was used to identify areas with adjacent forest-agriculture boundaries with a continuous extent of land cover on either side of the boundary.Transects were created across boundaries in the east/west direction at a ∼90°angle.The transects were approximately 1 km long (i.e., commensurate with the boundary layer height), with points every 10 m.In total, 44 transects were created in the study area (Figure S6 in Supporting Information S1).We then extracted T s values from mid-summer clear-sky scenes, all of which were from summer 2018, from the Landsat Provisional Surface Earth's Future 10.1029/2023EF003663 Temperature Product (Cook, 2014;Cook et al., 2014).Only one T s spatial profile was extracted for each transect, even if multiple images were available.The spatial resolution of Landsat imagery is 30 m, so the 10 m transects include several points within the same pixel.A smooth transition suggests that biophysical feedbacks on T s are likely linked to T a through advection and the formation of an internal boundary layer (Hsieh & Katul, 2009), since the surface itself is not mixed.For the lines in Figure 4e, we used a hyperbolic tangent function fit to each side of the transects (forest and agriculture).Distances were scaled from-π to π for fitting, then back-transformed for plotting.

Impacts of Reforestation on Air Temperature Trends
To investigate the impact of local land-cover change on long-term air temperature trends, we used monthly air temperature data from 398 United States Historical Climate Network (USHCN) meteorological stations.Meteorological stations, which were relatively evenly distributed across the study area (Figure S7 in Supporting Information S1), allow us to associate locations with their land cover type compared to gridded temperature data.
We evaluated the effects of land cover on annual temperature trends from 1900 to 2010 by focusing on both maximum temperatures during the growing season (June, July, August) and annual average temperatures.We did not conduct any interpolation of missing values.
We classified the weather stations based on the predominant land cover within a 500 m circular buffer surrounding each station, using 250 m FORE-SCE (FOREcasting SCEnarios of Land-use Change) backcasting grids from the US Land Cover Trends project (Sohl et al., 2007).This buffer size was chosen after preliminary investigation that suggested the cooler surface T s of forests could influence T a of nearby non-forest up to this range.This early assessment was subsequently refined and is presented in Figure 4e.A sensitivity analysis of the influence of different buffer sizes on temperature trends (from 100 to 1,200 m) supported the choice of a 500 m buffer, which optimally balanced capturing the impacts of reforestation on air temperature with minimizing confounding effects of unrelated processes (Figure S8 in Supporting Information S1).
The classification was based on three time points within the range of available data (annual from 1938 to 1992), 1938, 1965, and 1992.The intermediate year, 1965, helped exclude sites that had undergone multiple conversions throughout the 20th century.Pixels were categorized as reforestation areas (change from agriculture to forest) under two conditions: (a) if they were agricultural in 1938, converted to forest cover by 1965, and remained forested in 1992; or (b) they were agricultural in both 1938 and 1965, and then transitioned to forest by 1992.The areas predominantly surrounded by agriculture-to-forest conversion within 500 m of the weather stations were defined as "reforest" sites.Here, "predominant" land cover change refers to the land cover type with the greatest cumulative area, determined by summing the weights of each land cover type within the 500m buffer.Out of the 398 USHCN stations, 132 were predominantly surrounded by forest, 196 by agriculture, and 22 by reforestation.The remaining 48 sites were excluded due to multiple land cover transitions or deforestation in the 20th century.
To explore the impacts of reforestation on air temperature trends, we compared USHCN sites predominantly surrounded by reforestation to sites where the land cover remained stable as either agriculture or forest ("nonreforest") within a 50 km radius of the "reforested" site.For each year, we calculated the temperature difference between the reforest site and non-reforest site (or sites) within the 50 km buffer.This approach builds upon the paired-site flux tower analyses described in Section 2.2.1 by extending our understanding of the effects of reforestation on surface temperature to near-surface air temperature, covering a much longer period (back to the early 20th century), and including many more site "pairs."Our comparison of neighboring reforested and nonreforested site pairs included analysis of both growing-season maximum temperatures and average annual temperatures.We excluded sites that had more than 5 years of missing data from 1900 to 2010.We had a total of 22 reforest sites and 44 non-reforest sites within 50 km of the reforest sites, resulting in 44 comparisons for annual average data and 42 comparisons for growing season data.The lower number of growing season comparisons is due to missing data for two sites.The reforest sites had a median of two non-reforest sites within a 50 km radius.We based the 50 km radius on a previous study that showed changes in maximum air temperature up to 50 km away from the site of land cover change (Cohn et al., 2019).

Forest Age and Land-Cover Change Analyses
Monthly T a timeseries with a spatial resolution of 0.5°were obtained from the University of Delaware Air Temperature & Precipitation Data set (Willmott & Matsuura, 2015) provided by the National Oceanic and Atmospheric Administration's Oceanic and Atmospheric Research Earth System Research Laboratory Physical Sciences Division (NOAA/OAR/ESRL PSD), in Boulder, Colorado, USA, from https://psl.noaa.gov.The gridded data are interpolated weather station data (Willmott & Matsuura, 1995), primarily from GHCN2 (Global Historical Climatology Network) observations and the GSOD (Global Surface Summary of Day) archive.
Monthly mean air temperature from V4.01 of the University of Delaware Air Temperature & Precipitation Data set was used to assess observed long-term changes in T a across the continental United States.Per-pixel temperature change in T a was estimated using a season-trend model (function "STM") in the "greenbrown" package (Forkel & Wutzler, 2015) for R (R Core Team, 2020).The season-trend model in "greenbrown" is based on the additive decomposition model described elsewhere (Verbesselt et al., 2010(Verbesselt et al., , 2012)).Harmonic and linear terms are used to model seasonal variation and trend, respectively, effectively "detrending" the time series in a single step (Verbesselt et al., 2012).
To assess forest age, we obtained continental forest age maps from the North American Carbon Program (NACP), produced for the year 2006 (Y.Pan et al., 2012).Data were updated to provide forest ages for 2019 by adding 13 years to all forest ages.This approach assumes that all forests continued to regrow and were not cut between 2006 (when the product was produced) and 2019.Although there are certainly locations for which this assumption will not hold, they generally represent a small fraction of the land surface and are mostly contained in the southeastern US where pine plantations are routinely harvested and replanted (Carman, 2013; Figure S1 in Supporting Information S1).Furthermore, this method does not account for disturbances such as fires or insect infestations that could have affected stand age.Nonetheless, the low standard deviations of forest agescorresponding to the 2006 age map and typically around 10 years (Figure S1C in Supporting Information S1) -suggest a disturbance regime with limited spatial heterogeneity.Recent studies corroborate that forest loss in this region is predominantly not due to fires (Hansen et al., 2013;Tyukavina et al., 2022;van Wees et al., 2021).In the 20th century, fire frequency in the Eastern United States was substantially reduced when compared to frequencies before European settlement (Nowacki & Abrams, 2008).
To assess change in forest status (Figure 1b), land-cover change between 1938 and 1992 was calculated from 250 m FORE-SCE backcasting grids described in Section 2.4.1.Pixels were coded based on their land cover (cropland or forest), and per-pixel change was calculated between 1938 and 1992.To exclude sites that had undergone multiple conversions in the 20th century, an intermediate year, 1965, was also used.Pixels were classified into three categories: reforestation (change from cropland to forest), deforestation (forest to cropland), or "no change" (land cover type was consistent in 1938, 1965, and 1992).

Exploring the Links Between Forest Age and Air Temperature Change
To further investigate the influence of reforestation with a focus on the warming hole, we examined the relation between forest age and recent growing season trends in T a at the regional scale (Figure 5d).To align the recent "snapshot" of forest age with T a , we compared the 1 km NACP-derived forest age maps with recent trends  in T a from the University of Delaware climatologies (Willmott & Matsuura, 2015).We used the seasontrend approach described in Section 2.5.1 to estimate the slope of trends in annual T a time-series from 1970 to 2017 (Figure 5d).The slope of the temperature change from 1970 to 2017, ∆T a , was calculated for each pixel, and forest age data were aggregated to the coarser 0.5-degree scale of the T a data (Figure S1B in Supporting Information S1) using a mode function, "modal," in the "raster" R package (Hijmans, 2021), and then resampled using nearest neighbor interpolation.Then, we calculated focal correlations between aggregated forest age (Figure S1B in Supporting Information S1) and temperature trends, specifically, a moving window correlation between ∆T a from 1970-present and forest age in a 5 × 5 window using the "raster correlation" function in the "SpatialEco" R package (Evans, 2021).
We note that although gridded daily microscale (1 km) T a products such as Daymet should be relatively insensitive to local impacts of land cover on T a (Figure S9 in Supporting Information S1), the fingerprint of reforestation may be detectable from coarser yearly mesoscale T a estimates at 0.5-degree resolution.Observational and  modeling studies support regional-scale impacts of land cover change on mesoscale T a (Bonan, 2001;Mahmood et al., 2014), providing further justification for our approach.

Effects of Land Cover on T s : Local and Regional Biophysical Analysis
Across the study area, the difference between T s and the T a,Daymet was more negative for forests than grasslands and croplands most of the time (Figure 3).Since the T a, Daymet normalizes for macro-scale temperature variability, this result implies that forest surfaces are cooler than the surfaces of nearby grasslands and croplands by the same amount.The effect was most pronounced during the growing season, when forests were cooler than non-forests by 0.6-2.5°C(Figures 3a and 3b), with smaller reductions observed in spring and fall.
Next, we leveraged rich surface energy balance information from eddy covariance flux towers, beginning with six co-located ("paired") forest and grassland sites in the study region (Table S1 in Supporting Information S1; Zhang et al., 2020).Across these paired sites, forest T s was 4-5°C cooler, on average, than nearby grasslands during midday periods (Figures 3c and 3d), driven primarily by enhanced evapotranspiration in summer and enhanced sensible heat flux in winter that outweighed albedo-driven warming effects in the darker forests (Zhang et al., 2020).
Although the paired-site approach is well established for understanding the biophysical impacts of land-cover change (Zhang et al., 2020), only a handful of forest-grassland flux tower pairs exist.To expand the scope of inference, we synthesized T s observations from 58 Ameriflux EC tower sites across the EUS, again correcting for macro-scale climate variability with T a,Daymet (Table S2 in Supporting Information S1).The results revealed widespread daytime surface cooling in forests compared to non-forests (Figure 4a).Throughout the growing season, the mean difference between the tower-derived T s and the reference T a,Daymet was 5.7°C lower for tall (>20 m) forests compared to grasslands, and 4.6°C lower for short (<20 m) forests relative to grasslands (Figure 4a).When comparing forests to croplands, the differences were less pronounced, though tall (>20 m) forests were still relatively cooler than croplands.Specifically, T s (Figure 4a) was 8.1 ± 3.4 for grasslands, 4.1 ± 2.2 for croplands, 3.5 ± 2.1 for short forests, and 2.4 ± 1.3 for tall forests (all values are mean ± SD).

Effects of Land Cover on Near-Surface T a : Local and Regional Biophysical Analysis
For the surface cooling to extend beyond the stand-scale and thus contribute to the warming hole, changes in T s must translate to changes in near-surface T a .
We adopted several new or emerging approaches to quantify the extent to which the forest cooling effects on T s extend to T a .First, we harnessed flux tower data to estimate metrics of T a that are less sensitive to canopy effects (Novick & Katul, 2020): T aero and T extrap (Table 1).In the paired sites and across the regional network of flux towers, the midday growing season T aero and T extrap were cooler for forests than for grasslands (Figures 3d, 3e, and 4a-4d).
Consistent with expectations, forests had the strongest cooling effect on surface temperatures (Figure 4a), and smaller effects on air temperatures (Figures 4b-4d).When analyzing the air temperature metrics, T aero and T extrap were similar between forests and croplands, although T aero was substantially lower for tall forest stands (Figure 4b).T aero (Figure 4b) was observed to be 3.9 ± 2.1 for grasslands, 1.5 ± 1.6 for croplands, 1 ± 1.6 for short forests, and notably, 0.2 ± 0.9 for tall forests.Similarly, T extrap (Figure 4c) was 1.3 ± 0.8 for grasslands, 0.6 ± 0.84 for croplands, 0.6 ± 1 for short forests, and 0.1 ± 1.8 for tall forests.T extrap results showed tall forests to be 1.4°C cooler than grasslands, and short forests were 0.7°C cooler.
Despite the confounding influence of canopy effects on near-surface T a profiles, tower-measured T a also suggests a cooling effect of forests, although differences across biomes were smaller (Figure 4d).The mean of the difference between T s and tower-measured T a was 0.5 ± 0.7 for grasslands, 0.8 ± 0.4 for croplands, 0.9 ± 0.6 for short forests, and 1.5 ± 0.6 for tall forests.Throughout the growing season, the mean difference between the tower-derived T s and the tower-measured T a was 1°C cooler for tall forests compared to grasslands, and 0.7°C cooler for tall forests relative to croplands (Figure 4d).
Next, to provide an independent perspective on surface and air temperature coupling, we used high-resolution (30 m) Landsat T s retrievals to evaluate the extent to which transitions in T s at forest-cropland boundaries were smooth or abrupt (Figure 4e; Figure S6 in Supporting Information S1).A relatively smooth temperature transition from cooler forests to warmer croplands was observed (Figure 4e) that extends over length scales of several hundred meters.

Long-Term Meteorological Observations and Signatures of Land-Cover Change: Local and Regional Historical Analysis
To evaluate the signatures of land-cover change in long-term meteorological observations, we compared longterm records of air temperature from weather station sites, estimates of forest age, and gridded air temperature data (see methods Section 2.4.1).First, long-term temperature trends from 398 USHCN sites in the study region were examined.The examination showed that temperatures remained relatively stable throughout the 20th century, with a decline in average temperatures in the late 1950s (Figure 5a).Next, we investigated the cooling effect of regrowing forests by analyzing the difference between T s and the reference T a,Daymet as a function of forest age.Results indicated that the cooling effect was strongest for forests around 20-40 years old, as the difference reached its lowest point at that age (Figure 5c).
To further explore the cooling effect of reforestation, we compared 10-year rolling means of T a from historical USHCN stations in sites predominantly surrounded by reforestation ("reforest") and neighboring sites that did not undergo land cover conversion ("non-reforest") within 50 km.In terms of average annual temperatures, reforested sites were consistently cooler than their non-reforesting counterparts throughout the 20th century (Figure 5b, blue line).The results were more nuanced for maximum growing season temperatures.Sites with nearby reforestation tended to be warmer in the early half of the 20th century and cooler in the latter half, and the magnitude of this cooling effect increased throughout the study period.By the end of the 20th century, reforesting sites were up to 1°C cooler than their non-reforesting neighbors in terms of maximum growing season air temperatures (Figure 5b, red line).
Finally, to explore the relation between forest age and temperature trends, we explored spatial variation in the correlation between decadal-scale growing season T a trends (from University of Delaware 0.5°monthly climatologies) and forest age across the continental US (Figure 5d).A positive correlation coefficient indicates that pixels dominated by younger forests (i.e., <100 years as of 2019) experienced less warming during this period than pixels dominated by more mature stands (see inset to Figure 5d).This correlation coefficient was positive across much of the study region on the annual timescale (63% of pixels; Figure 5d), and when considering growing season temperature trends alone (Figure S10B in Supporting Information S1), consistent with the expectation that reforesting areas experience less warming.However, there is substantial variability across the study region, with negative relations between forest age and ∆T a observed, including in the central Appalachians and Ozarks.

Discussion
The study provides compelling evidence of the biophysical climate benefits of reforestation in the EUS, while also establishing a clear relation between land-cover changes and temperature shifts observed throughout the 20th century.Taken together, these findings indicate reforestation has a cooling effect on surface and near-surface air temperature in the EUS, and likely contributed to the slower pace of warming in the region.Both ground-and satellite-based observations indicate that EUS forests cool the land surface by 1-2°C annually (Figure 3) compared to nearby surfaces with short-stature vegetation.During midday in the growing season, surface cooling Earth's Future 10.1029/2023EF003663 is 2-5°C (Figures 3c and 4a), and forests aged 25-50 years exhibit the strongest cooling effect (Figure 5c).The surface cooling extends to the near-surface air, with forests reducing midday growing-season air temperature by up to 1°C (Figures 4d and 5b).Historical weather station data and regional-scale gridded long-term air temperature data analyses establish a link between reforestation and the observed lack of warming in the EUS (Figure 5).Weather stations located near reforesting areas recorded temperatures that were 0.5-1.0°Ccooler than stations surrounded by land that experienced little change in forest cover.Overall, these results highlight the substantial adaptation potential of reforestation as a nature-based climate solution.
However, these findings may not be applicable across all temperate regions or for all strategies, such as afforestation.The implementation of NbCS must be undertaken with consideration to avoid unintended consequences that could lead to net warming (Novick, Metzger, et al., 2022;Peng et al., 2014).For instance, while reforestation in boreal regions may seem beneficial, changes in albedo and evaporation dynamics could lead to warming in the long-term (Liu et al., 2019;Randerson et al., 2006).Moreover, potential disturbances-such as climatic stressors, wildfires, and insects-carry substantial implications for the success of NbCS.These disturbances could compromise forest carbon storage (Anderegg et al., 2022, particularly in regions more susceptible to these risks like the Western United States), even in regions such as the Eastern United States where these are historically less prevalent.Although forests in the EUS are generally less prone to such disturbances compared to those in the West (Barbero et al., 2015;Bentz et al., 2010), it is nonetheless important to recognize these risks when evaluating region-specific NbCS potential.

Assessing Land Cover Impacts on Surface and Near-Surface Air Temperature
Assessing the direct consequence of land-cover change on surface temperature is less ambiguous, given the abundance of satellite observations of T s and complementary measurements of surface energy fluxes from ground-based towers.We have known for some time that reforestation tends to increase T s in the boreal zone (Bright et al., 2017;Lee et al., 2011;Swann et al., 2010) and decreases T s in the tropics (Bonan, 2008).In this study of temperate ecosystems, we demonstrated that forests in mid-latitude of the eastern US can contribute to a T s reduction of 2-5°C at midday during the growing season (Figures 3 and 4a).Furthermore, these reductions in T s can be achieved within approximately 20 years of forest regeneration (Figure 5c) and remain stable across mature forest classes (Figure 4a).Our findings, combined with other recent work on the topic, contribute to a growing consensus that the presence of forests tends to have a direct cooling effect on T s across much of the temperate zone.
Remote sensing data inherently favor clear-sky conditions, introducing potential bias into the results presented in Figures 3a and 3b.This bias, however, is mitigated considerably by conducting similar comparisons with flux data (Figure 4a), which are not subject to these clear-sky limitations.Another potential source of bias is associated with the timing of the MODIS overpass.Satellite retrievals are obtained at approximately 1:30 p.m. local time, which could be earlier than when Daymet would record the daily maximum air temperature, particularly in summer.This mismatch might cause the difference between T s and the Daymet T a to be lower than it would be in the absence of a temporal discrepancy.Our study, however, primarily focused on comparing the T s -T a difference for forested and non-forested areas.Consequently, even if the timing of the MODIS overpass does not align with the peak daily temperatures estimated by Daymet, any small discrepancy would largely be removed when analyzing their difference.Further supporting this point, we found differences in T s -T a between forest and nonforest flux sites (Figure 4a) of magnitudes similar magnitude to those from derived from the remote sensing data.These complementary findings lend further credence to our comparative analysis between forested and nonforested areas, thereby strengthening our confidence in the overall results, despite the potential sources of bias.
Assessing the impact of land cover change on near-surface air temperature is more challenging due to technical and methodological limitations.Yet, understanding the potential of reforestation and other natural climate solutions to confer climate adaptation benefits requires that we quantify land cover impacts on near-surface air temperature.Furthermore, for the surface cooling to extend beyond the stand scale and thus contribute to the warming hole, changes in T s must translate to changes in T a.In turn, evaluating the impacts of land cover on nearsurface air temperatures requires overcoming technical and methodological challenges, including data scarcity and canopy structural effects that prevent straightforward comparisons of air temperature measured above different ecosystems.We adopted several new or emerging approaches to overcome the technical and methodological challenges associated with evaluating T a .
First, we harnessed flux tower data to estimate metrics of T a that are less sensitive to canopy effects.By relying on proxies for near-surface air temperature that are relatively insensitive to canopy structural effects (e.g., T aero and T extrap ), we demonstrate that, across much of the study region, the cooling effect of forests on T s extends to the near-surface air temperature above the canopy (Figures 3c and 3d; Figure 4).Forest cover impacts on T a are smaller than impacts on T s (∼1°C for midday growing season periods, Figures 3d and 4) but still consequential, particularly compared to the magnitude of historic and predicted changes in T a due to climate change.
Next, to provide an independent perspective on surface and air temperature coupling, we used high-resolution (30 m) Landsat T s retrievals to evaluate the extent to which transitions in T s at forest-cropland boundaries were smooth or abrupt.A smoother transition suggests that biophysical feedbacks on T s are linked to T a through the formation of an internal boundary layer, assuming the thermal inertia of leaves is small.A relatively smooth temperature transition from cooler forests to warmer croplands was observed (Figure 4e) that extends over length scales typical of adjustment distances needed for the equilibration of the air temperature internal boundary layer.Furthermore, the smooth transition of T s across forest-grassland boundaries implies coupling between T s and T a that extends at least several hundred meters beyond the ecosystem boundary.This result helps to confirm previous theoretical modeling (Hsieh & Katul, 2009;Li & Wang, 2019) and augments a similar exercise in the tropics (Cohn et al., 2019) by documenting these processes at a large spatial scale over much of the EUS.Importantly, it is over this distance that biophysical impacts on T a are substantial enough to feedback on T s .It is likely that reforestation impacts on T a extend further, but beyond a few hundred meters are not great enough to drive significant changes in grassland T s .
Our analyses have shown that forests in the EUS exert a cooling effect on both surface and air temperatures, and that this effect can extend across ecosystem boundaries.Importantly, this cooling benefit is most pronounced during midday summer periods (Figures 3c and 3d; Figure 5b), which are typically associated with high heat stress and extreme events.These findings suggest that, in temperate zones, reforestation may provide the greatest climate adaptation benefit precisely when it is most needed.

Linking Reforestation to the Observed Lack of Warming in the Eastern US
To establish a link between reforestation and the observed lack of warming in the EUS, we analyzed historical weather station data to associate near-surface air temperature records with land cover changes during the 20th century.Additionally, we conducted a regional-scale analysis to investigate the relationship between spatial patterns of reforestation and temperature patterns in the EUS.
Overall, our analysis of long-term T a records from 398 USHCN weather stations is consistent with the general understanding of 20th temperature trends in the EUS as it reveals no overall warming trend (Figures 1c and 5a).We observed a sharp decrease in average temperatures in the 1950s, which corresponds with previous studies that found an abrupt climactic regime shift in 1957-1958 in the EUS (Partridge et al., 2018;Rogers, 2013).The causes of this abrupt, uniform cooling are likely multifaceted.The abrupt cooling could be related to changes in jet stream position, specifically sharp decreases in the Meridional Circulation Index (MCI; Tosca et al., 2017).However, the impact of decreases in the MCI is greater on winter temperatures, which decreased more uniformly during the twentieth century than summer temperatures (Figure S11 in Supporting Information S1).Therefore, focusing on summer temperature time-series in addition to annual temperature time-series can help to distinguish the influence of the MCI from other mechanisms, including reforestation.
We investigated the signatures of reforestation in long-term climate data by analyzing the difference in longterm T a records between weather stations predominantly surrounded by reforestation and nearby (within 50 km) weather stations that did not undergo land cover change (remained agriculture or forest) throughout the 20th century.In the early 20th century, we found that sites with nearby reforestation tended to be warmer than sites without reforestation in terms of maximum daily growing season temperatures (Figure 5b).This pattern is consistent with expectations if these areas were sparsely vegetated at the time (Figure 1), likely being abandoned agricultural or marginal lands with limited vegetation.As a result, these areas would have experienced lower rates of transpiration prior to reforestation, compared to their levels after reforestation.However, as the 20th century progressed, sites predominantly surrounded by reforestation became increasingly cool relative to their non-reforesting neighbors within 50 km.By the end of the 20th century, reforesting sites were up to 1°C cooler than their non-reforesting neighbors in terms of maximum growing season air temperatures (Figure 5b).The magnitude of the cooling effect is consistent with the results from the tower air temperature comparisons (Figures 3c,3d and 4), providing independent evidence of the impact of reforestation on near-surface air temperature.
Analysis of the forest cooling by age indicated that the cooling effect of regrowing forests takes about 20 years to fully develop, with forests around 20-40 years old exhibiting the strongest cooling effect (Figure 5c).This time frame is consistent with the time it takes for regenerating forests in the region to achieve levels of hydrological function comparable to mature forests (Ford et al., 2011).To offer historical context, during the growing seasonwhen the cooling impact of regrowing forest is at its peak-reforesting sites were consistently cooler than nearby non-reforesting sites by the late 1950s (Figure 5b).This timeframe roughly corresponds to forests aged between 20 and 30 years, assuming the 1930s as a starting point due to widespread agricultural abandonment and federal reforestation efforts during that decade.The gap between maximum growing-season temperatures in forested versus nearby non-forested areas widened throughout the 20th century, supporting the idea that as forests grew, the summer cooling effect increased (Figure 5b).
Regional-scale analysis of gridded long-term air temperature data provided an independent, yet complementary approach to the weather station analyses.The analysis showed that younger forests were associated with lower historic rates of warming across most of the study region (Figure 5d).This finding, combined with the weather station analyses, suggests that reforestation had a cooling effect on near-surface air temperature across a wide swath of the EUS.Parsing the trends in long-term T a time series as a function of land cover and historic land-cover change (e.g., the results in Figure 5) captures both the influence of regional-scale non-local effects and the finerscale direct effects.However, we note studies focusing on 21st-century forest cover (e.g., Hansen et al., 2013) found considerable forest turnover in the EUS since 2000.If these findings accurately reflect forest dynamics, they could introduce complexities to the relationships presented in Figures 5d and 5c, which specifically explore the impact of forest age.Our methodology, employing aggregated data and spatial moving window analysis, is designed to mitigate the impact of these potential localized variations.Nonetheless, we recognize the potential for biases in our results related to forest age, particularly if the actual average stand age is systematically underestimated.We estimate that the potential impact of such a discrepancy on the conclusions drawn from our forestage related findings is likely minimal.Even a considerable reduction in stand age (10 years) across a substantial portion (20%) of the study area would not alter the conclusions derived from Figures 5c and 5d (Figure S12 in Supporting Information S1).
While uncertainties remain regarding the characterization of forest age and historical land cover, our results support the expectation that 20th-century EUS reforestation had a net cooling effect that extended well beyond the surface and local stand-scale.For weather stations located near to, but outside of, reforested areas, the effect amounts to a suppression of T a on the order of 0.5-1°C for the latter half of the 20th century.

Non-Local Effects of Land-Use Change
A more holistic perspective on the efficacy of nature-based climate solutions accounts for the possibility that local (e.g., ecosystem-scale) changes in land cover can initiate non-local impacts over much broader scales (Swann et al., 2012;Williams et al., 2021).For example, reforestation can lead to increased evapotranspiration, resulting in increased cloud cover and precipitation (Cerasoli et al., 2021;Manoli et al., 2016) that extend across the landscape.These effects would tend to amplify local cooling, particularly during the daytime.Localized landcover changes can also cause shifts in atmospheric circulation, which can have continental or even globalscale consequences for temperature, precipitation, cloudiness, and other meteorological drivers (Pongratz et al., 2010;Swann et al., 2012;Winckler et al., 2019).Earth system models have been the primary tool for exploring these teleconnections, with many studies suggesting a dramatic role for non-local processes to amplify or counteract the local impacts of land-cover change on surface and near-surface temperature.However, results from these modeling studies are sometimes contradictory, and the difference between model predictions and observations can be large (Bonan, 2008;De Hertog et al., 2022).Notably, such discrepancies can include underestimations of the biophysical climate impacts associated with the expansion of vegetation cover (Li et al., 2020).
This study employs observational approaches to indirectly assess the non-local effects of land-cover change.We investigate how surface temperature effects can extend to the air and be transported across the landscape, providing a generalizable method to investigate these connections.Our approach includes exploring gradients in surface temperature (T s ) across ecosystem boundaries, as shown in Figure 4, to uncover the potential local extent of such effects.This connection is crucial because long-term records of air temperature, which are used to delineate the warming hole, are made in open clearings.By uncovering how surface temperature effects extend to the air, we can gain insight into how local changes in land cover may initiate non-local impacts across much broader scales.It is important to note that these observational approaches have many limitations, as they cannot fully account for the possibility that non-local effects of land-use change within and outside the study region are influencing long-term temperature trends in the EUS.
For example, changes in agricultural management, such as agricultural intensification and increased irrigation use, are known to have a local cooling effect (Mueller et al., 2016), and many areas of the EUS have experienced these management shifts (Spangler et al., 2020).Changes in agricultural management that promote cooling could obscure the influence of reforestation in the comparison of long-term trends from forested and cropland ecosystems (e.g., Figure 5).Thus, the results presented here can be considered conservative.It is also possible that land-use changes occurring well outside of the study region may be driving widespread trends in long-term temperature trends in the EUS via teleconnections (Swann et al., 2018), or that certain non-local temperature effects of reforestation or deforestation may be opposite in sign to the local effects (De Hertog et al., 2022;Pongratz et al., 2021).These non-local teleconnections are currently only possible to explore with modeling studies that tend to rely on idealized experiments that force instantaneous and extreme changes in land cover across broad regions.Evaluating all modes of historical land-use change that might have affected climate in the EUS would be computationally expensive (if not impossible) and would still be unlikely to resolve, with precision, the relatively small changes in T s and T a that are revealed by the observation-driven approach taken here.

Conclusions
Various hypotheses have been proposed to explain the observed lack of 20th-century warming in the eastern United States (e.g., Meehl et al., 2012;Z. Pan et al., 2004;Partridge et al., 2018;Tosca et al., 2017).The work here does not identify widespread reforestation as the sole factor causing the EUS warming hole or its trend, but multiple independent data sources suggest it can be an important contributor to this lack of historic regional warming.Beyond that, the study provides robust evidence of local biophysical climate benefits of reforestation in the EUS.The strong and persistent increase in forest cover throughout the region in the 20th century contributed to cooling, which is consistent with observed temperature changes.In addition, the findings demonstrate that reforestation has a consistent cooling effect on both surface and air temperatures, especially during midsummer periods when high temperatures can be most harmful.These findings emphasize the potential for reforestation to provide local climate adaptation benefits in temperate regions such as the EUS, highlighting the importance of biophysical co-benefits of nature-based climate solutions.
• Description: Site descriptions and details on eddy covariance data processing for the six paired flux sites using the REddyProc processing tool (Wutzler et al., 2018).• Availability: The data used in this study can be found in Zhang et al. (2020).4. Temperature Extrapolation Model (Novick & Katul, 2020) • Description: Results for temperature extrapolation into the first 10 m of the surface layer after conceptual flattening of ecosystems.• Availability: Details and code for the temperature extrapolation model can be found in Novick and Katul (2020).5. National Ecological Observatory Network (NEON) Towers (Metzger et al., 2019) • Description: Towers installed and operated by NEON for ecological observations.
• Availability: The MODIS data set used in this study is available in Wan et al. (2015).8.U.S. Geological Survey North American Land-Cover Characteristics Data set (EROS, 2017) [Dataset] • Description: Land-cover characteristics data set.
• Availability: The land-cover data set used in this study was provided by the U.S. Geological Survey's North American Land-Cover Characteristics project (EROS, 2017).9. National Land Cover Database (NLCD) (Dewitz, 2019) [Dataset] • Description: Land cover data set used to identify forest-agriculture boundaries.
• Availability: The National Land Cover Database (NLCD) data set used in this study can be accessed through Dewitz (2019).10.Landsat Provisional Surface Temperature Product (Cook et al., 2014) • Description: Landsat-based surface temperature product.
• Availability: The Landsat Provisional Surface Temperature Product used in this study is described in Cook et al. (2014) and can be accessed here: https://www.usgs.gov/landsat-missions/landsat-collection-1us-analysis-ready-data.11.FORE-SCE (Sohl, 2018) • Description: FOREcasting SCEnarios of Land-use Change backcasting grids from the US Land Cover Trends project.• Availability: The FORE-SCE grids used in this study are described in Sohl et al. (2007) and can be accessed through Sohl, 2018.12. United States Historical Climate Network (USHCN) Meteorological Stations • Description: Meteorological station data for the United States.
• Availability: The University of Delaware Air Temperature and Precipitation Data set used in this study is provided by Willmott and Matsuura (2015) through NOAA/OAR/ESRL PSD.14. "greenbrown" R Package (Forkel & Wutzler, 2015) [Software] • Description: R package used for season-trend model estimation.
• Description: R package with the "modal" function used in the analysis.
• Availability: The "raster" R package with function is available through Hijmans (2021).

Figure 1 .
Figure 1.The Southeastern United States' warming hole' and corresponding forest status.(a) Forest age estimates (1 km) as of 2019 calculated from forest age data from the North American Carbon Program.(b) Land conversion between agricultural land and forests from 1938 to 1992 calculated from 1 km FORE-SCE backcasting grids from the US Land Cover Trends project.The bounding box indicates the study area, and details on data sources are provided in the methods.(c) Trend in temperature change from 1900 to 2010 (∆T a , °C/50 years) calculated using a season-trend model applied to 0.5°T a grids from University of Delaware monthly climatologies provided by the NOAA.(d) Historical photo of 15-person planting crew on 05/09/1932 in Tucker County, West Virginia.Original photo at the Forest Service Office in Elinks, WV.Obtained from https://www.loc.gov/pictures/resource/hhh.wv0307.photos.041150p/.

Figure 2 .
Figure 2. Conceptual representation of the methodological approaches employed in this study.The diagram is divided into four quadrants to differentiate local and regional scales and biophysical and historical contexts.Each quadrant outlines the associated methods and their purpose.The methodologies span a broad range of space (local to regional) and time (diurnal to long-term historical trends) scales and enriches biophysical analyses (energy-radiation balance combined with physiological and aerodynamic approaches) with empirical or data-driven approaches.

Figure 3 .
Figure 3.A forest surface cooling effect is evident in both satellite and flux tower (a)-(b) observations (c), (d).(a) Average difference between daily ∼1:30 p.m. surface temperature (T s ) and daily maximum air temperature (T a,Daymet ) for 2003-2018 for forests (top row) and combined grasslands and croplands ("other," second row).Negative values indicate cooler surface than air temperature (surface cooling) and positive values indicate warmer surface air temperature.(b) The seasonal cycle of T s -T a,Daymet for forests (blue), grasslands (yellow), and croplands (orange) for the study region (bolded lines), with latitude ranges indicated by faint dashed lines.(c), (d) Diurnal time series of the difference between forest and grassland T s (blue), T aero (black), T extrap (gray), and tower-measured T a (yellow) for six eddy covariance site pairs (TableS1in Supporting Information S1) for the growing season (c) and the dormant season (d).

Figure 4 .
Figure 4. Extension of surface cooling to the near-surface air.(a-d) Difference in tower-measured or derived temperature metrics and reference T a,Daymet .Each boxplot shows the monthly midday growing season site-level means, with forested ecosystems split by canopy height (≤20 m or >20 m).Horizontal lines indicate the median, boxes indicate the interquartile range, whiskers indicate the data range (=1.5 times the interquartile range), and the symbols "+" indicate outliers.Letters indicate significant differences between groups evaluated using a two-sample t-test at a significance level of p = 0.05.Note the y-axis range varies between panels.Each panel is informed by data from 13 grass towers, 19 cropland towers, 10 towers in forests <20 m tall, and 16 towers in forests >20 m tall.(E) Horizontal profiles ("transects") of T s across the forest:cropland boundary.Values of ΔT s (sampled every 10 m) are relative to the temperature at the forest:cropland boundary.Negative values of ΔT s indicate lower temperatures than at the boundary, and positive values of ΔT s indicate higher temperatures than at the boundary.Points left of the zero-distance line indicate forest (blue), and points right of the zero-distance line indicate agricultural croplands (orange).The blue and orange lines indicate a hyperbolic tangent function fit to the pooled transect data.

Figure 5 .
Figure5.Impact of land cover and forest age on long-term temperature trends.(a) Average trend in pooled annual T a anomalies for 398 USHCN sites in the EUS.The blue line denotes the 10-year moving average.(b) Difference in T a between reforested and non-reforested USCHN sites within a 50 km radius.Red and blue lines depict smoothed summer maximum and annual average temperatures, respectively.Negative values indicate cooler temperatures at the reforesting site than its neighboring non-reforesting site(s).(c) July temperature difference (T s -T a,Daymet ) versus forest age, using data from 30,000 randomly selected MODIS pixels across the study area.The blue line represents the Generalized Additive Model smooth function, while the surrounding white area depicts its 99% confidence interval.(d) Spatial moving window correlation (5 × 5 window) between forest age and recent long-term T a trends.A negative correlation indicates older forests are associated with greater cooling, and a positive correlation indicates younger forests are associated with greater cooling (inset).

Table 1
Summary and Brief Description of the Relevant Temperature Metrics Used in This Study BARNES ET AL.