Biophysical feedback of forest canopy height on land surface temperature over contiguous United States

Forests are considered important in the mitigation of climate change. Biophysical effects of afforestation and deforestation on land surface temperature (LST) have been extensively documented. As a fundamental variable of forest structure, however, few studies have investigated the biophysical feedback of forest canopy height (FCH) changes on LST at large scale. This study is designed to investigate the impact of FCH changes on local land LST and clarify the biophysical processes controlling LST change from 2003 to 2005 over the contiguous United States, based on satellite observations. To this end, one satellite-based FCH product is selected, and the space-for-time approach, together with the energy balance equation, is applied. Results show that for different forest types, namely evergreen forest (EF), deciduous forest (DF), and mixed forest (MF), taller forests present a greater net cooling effect (0.056–0.448 K) than shorter forests at annual scale. The increase in net radiation and sensible heat flux was less than the increase in the latent heat flux when FCH classes converted from shorter to taller, resulting in annual net cooling effects. Furthermore, the cooling effect of EF is stronger than that of DF and MF, whether for tall, medium, or short FCH classes. Multiple regression analysis reveals that the changes in biophysical components can effectively explain the LST change during the growing season. Our findings provide a new insight for forest management decision-making with the purpose of mitigating climate warming.


Introduction
Through both biogeochemical and biophysical processes, forests play a crucial part in mitigating climate change (Arora and Boer 2010, Liao et al 2020). The biogeochemical process regulates climate by assimilating carbon dioxide (CO 2 ) absorbed from the atmosphere and storing it as carbohydrate in tree biomass and forest soils while the biophysical process influences climate through altering surface biophysical properties and modifying the exchange of energy and moisture between the land and atmosphere (Carvalhais et al 2014, Bright et al 2017, Abera et al 2019, Su et al 2021. In recent decades, research on the forest biophysical effect has received more attention than the biogeochemical effect due to its high spatial heterogeneity and complexity (Zhang and Liang 2018).
Forest canopy height (FCH) is acknowledged as a fundamental attribute of forest structure and primary driver of forest above-ground biomass, species diversity, and forest functions (Zhang et al 2014, Tao et al 2016, Xu et al 2018a, Potapov et al 2021. Investigating the biophysical effect of FCH changes is of great importance due to (a) forest height structures being continuously influenced by various natural and anthropogenic factors such as fires, water stress, deforestation, afforestation, as well as wood thinning and harvesting all around the globe (Koch et al 2004, Amiro et al 2006 van Mantgem et al 2009, Tao et al 2016; (b) the FCH changes significantly affecting the surface biophysical properties (Kumkar et al 2020), including albedo (primarily controlling the fraction of solar energy absorbed by surface), aerodynamic roughness (primarily regulated by the surface roughness length), and surface resistance (primarily regulated by water ability and vegetation structure), and hence directly impacting the partitioning of net radiation into latent and sensible heat fluxes and ultimately surface land surface temperature (LST) (Kuusinen et al 2014, Xu et al 2018b, Moon et al 2020. Therefore, the evaluation of the biophysical climate effects with different FCHs contributes to a better understanding of the feedback of forest on local climate. Nowadays, research studying forest biophysical climate impact mainly falls into two categories: biophysical effect with and without land use and land cover change (LULCC). The biophysical effect of LULCC (e.g. afforestation) on local, regional, and global climate has been extensively explored using in situ measurements, satellite observations, and climate modeling techniques (Cao et al 2019, Li et al 2020, Zhang et al 2020, Zeng et al 2021. These studies agreed that afforestation in boreal zones increases the surface temperature due to the stronger albedoinduced warming effect than evaporative cooling effect, but the opposite occurs in tropical zones as a result of the stronger cooling effect of evapotranspiration (ET) (Li et al 2015). In temperate zones, the biophysical climate effect of afforestation is still ambiguous depending on the competing effects of albedo and ET (Huang et al 2018). Although the biophysical effect of land cover changes on local climate have been well studied, in contrast, there are still large knowledge gaps in relation to forest biophysical effects that do not involve land cover conversion.
Studies related to the biophysical impact of forests that do not involve LULCC focused primarily on forest management-induced changes in forest structural properties (e.g. FCH, leaf area index (LAI), and forest age) (Kumkar et al 2020, Zhang et al 2021). Changes in forest structure trigger significant changes in surface biophysical properties such as surface albedo, influencing surface energy balance, and hence the LST (Abera et al 2019). For example, Kumkar et al (2020) employed an offline model to simulate the impact of forest structure changes on local LST and attributed the contributions of biophysical components to LST change. They revealed that fully developed forests (i.e. highest LAI) decreased LST annually by 0.04 K due to a stronger evaporative cooling effect, whereas undeveloped forests (lowest LAI) increased LST by 0.14 K at annual scale owing to lower ET. Zhang et al (2021) investigated the biophysical effect of forest age changes on local LST using eddy covariance measurements from five paired Ameri-Flux evergreen needleleaf forest (ENF) sites. They demonstrated that older ENF has an annual net cooling effect of 1.7 K more than younger ENF.
FCH, as one of the most frequently used forest vertical structure variables, has received much attention regarding quantifying its affect on LST at city level (Zhang et al 2014, 2017, Yu et al 2018. Many in situ and remote sensing observational studies have demonstrated that LST is strongly negatively correlated with FCH (Yu et al 2018, Helletsgruber et al 2020, Vanneste et al 2020. However, these studies focused on urban or site scale. To our knowledge, few studies have been conducted to quantify the biophysical effect of forests with canopy height change on LST at broad geographical scales. Furthermore, the biophysical mechanisms of FCH change on LST have been rarely documented. Recent advances in the development of global FCH satellite products provide the possibility of evaluating the biophysical impact of FCH changes on local climate at national scale. In this study, we used satellite-derived products and a space-for-time approach with the objective of investigating the biophysical impacts of FCH changes on surface energy balance and LST. Specifically, we classified FCH into three categories (i.e. tall, medium, and short) and aimed to address the following three questions: (a) What is the LST response among different FCHs from 2003 to 2005 over the contiguous United States (CONUS)? (b) How do the associated surface energy fluxes and biophysical properties change? (c) Can the changes in biophysical components explain the LST change?

Remote sensing datasets
A large variety of remotely sensed datasets, including FCH, land cover, LST, albedo, emissivity, ET, surface incoming shortwave radiation, and elevation over CONUS were used in our study. Here, land cover was mapped using the 2016 National Land Cover Database (NLCD) product. Three different forest types including evergreen forest (EF), deciduous forest (DF), and mixed forest (MF) were extracted for further study. The distribution of EF, DF, and MF in CONUS is displayed in figure 1(a). The FCH product that we used is a global canopy height map with 1 km spatial resolution, which was developed by Simard et al (2011). In this study, we reclassify the FCH into less than 10 m, 10-20 m, and greater than 20 m to represent tall, medium, and short forest, respectively ( figure 1(b)). Furthermore, to avoid the randomness of this classification, we also set another three categories: (a) less than 5 m, 5-15 m, and greater than 15 m; (b) less than 5 m, 5-20 m, and greater than 20 m; (c) less than 15 m, 15-25 m, and greater than 25 m. The detailed descriptions of the datasets used in this study are provided in Text S1. All of the remote

Space-for-time strategy
A space-for-time approach (Yang et al 2020) was employed to quantify the impacts of FCH changes on local LST at 1 km resolution. The key point of this method is to compare the LST difference (∆LST) between tall forests and adjacent short forests. This method assumed that all adjacent tall forests and short forests pixels share a similar background climate so that local LST differences between tall forests and short forests can be attributed to FCH changes. Furthermore, since temperature is very sensitive to elevation variation, we controlled the elevation difference between adjacent tall forests and short forests within 50 m (Yang et al 2020). The process of this method includes three steps. Firstly, a square window with initial edge length (40 km) centered on a tall forest pixel is set. Secondly, we search for short forest pixels in this window. The short forest pixel with an elevation difference exceeding 50 m compared with central tall forests will be excluded. Thirdly, if the number of remaining short pixels in the window is larger than ten, the research will stop and ∆LST will be calculated. Otherwise, the edge length of the square window will increase step by step. The maximum edge length of the window can not be greater than 100 km. If so, the window will be discarded. The workflow of this method is provided in figure S1 (available online at stacks.iop.org/ERL/17/034002/mmedia). FCH was classified as tall, medium, and short in our study. Thus, the biophysical impact of FCH changes on LST was investigated using the following equation: where LST tall , LST medium , and LST short represent the surface temperature for tall, medium, and short forests, respectively. Negative ∆LST denotes a cooling effect of taller forests.

Biophysical controls on LST changes
We used an energy balance equation to further illustrate the biophysical process leading to ∆LST. The land surface energy balance equation can be expressed as follows: where R n , R ns , and R nl are net radiation, net shortwave radiation, and net longwave radiation, respectively. SW in , LW in , SW out , and LW out are incoming shortwave and longwave radiation and outgoing shortwave and longwave radiation, respectively. SW out and LW out can be further inferred as: where α is albedo, σ is the Stefan-Boltzmann constant (5.67 × 10 Moreover, the R n absorbed by land surface is approximately balanced by energy that is transferred out from land surface. Thus, R n can also be calculated as: where LE, H, and G are latent heat flux, sensible heat flux, and ground heat flux, respectively. . By combining (4)- (7), the land surface energy balance equation can be formalized as: As the space-for-time approach assumes that the adjacent tall forests and short forests share a similar background climate, the difference in SW in and LW in can be ignored (Ge et al 2019). Moreover, the contribution of G was unconspicuous at monthly and annual scales compared with other components (Purdy et al 2016, Zhang et al 2021). Therefore, the terms of the energy balance equation can be formulated as follows due to FCH changes where ∆(•) signifies the difference of each component between different FCH classes, for instance, ∆α = α tall − α short is the difference in α between tall forest and short forest. Similar to ∆LST, ∆α, ∆R ns , ∆LE, ∆H, and ∆R nl were also obtained using the space-for-time method. The formulation above is the changes of surface energy fluxes between tall and short forest. The changes in surface energy fluxes of tall and medium forest, and medium and short forest are defined similarly.

Feedback of FCH changes on LST
The spatial distributions of observed LST change showed that taller forests were generally cooler than shorter forests over CONUS at the annual scale, whether for EF, DF, or MF (figure 2). The probability density functions revealed that the pixels with negative values (cooling effect) were in the majority (figure 3). Black vertical lines represent ∆LST of 0 K and blue vertical lines represent the mean ∆LST. Figure 4 shows that for EF, the annual mean LST difference between tall and short forests (∆LST TS ) is −0.448 ± 0.004 K (mean ± 95% confidence interval, the same hereafter), between tall and medium forests (∆LST TM ) is −0.295 ± 0.002 K, and between medium and short forests ( For all forest types, the annual mean ∆LST TS was larger than ∆LST TM , indicating that medium forests were cooler than short forests. Moreover, the observed annual mean ∆LST MS also verified this result. The mild difference between ∆LST TS − ∆LST TM and ∆LST MS can be attributed to the discrepancy of pixels in the square window in the space-for-time approach. Overall, at annual scale, tall forests (EF, DF, and MF) produced the largest cooling effect, followed by medium and short forests. This conclusion was also applied to the growing season. However, for DF in the non-growing season, conversion from short DF to tall or medium DF led to a mild warming effect. The reason was detailed in section 4. Additionally, the cooling effect of taller forests was significantly higher during the growing season than the non-growing season. Specifically, conversion from short forests to tall forests obtained the highest mean ∆LST for EF (−0.876 ± 0.006 K), DF (−0.449 ± 0.002 K), and MF (−0.463 ± 0.005 K) in the growing season. For different forest types, the cooling effect of EF is stronger than that of DF and MF, whether for tall, medium, or short FCH classes. The above results were based on the FCH of less than 10 m, 10-20 m, and greater than 20 m. The impact of FCH changes on LST based on the other three categories were displayed in figures S2-S4. In general, these results demonstrated that taller forests had a stronger cooling effect than shorter forests.
To further illustrate the impact of FCH changes on LST, the relationship between difference (∆FCH) and ∆LST was established using correlation analysis    (figure 5). As with ∆LST, ∆FCH was also calculated in a square window. The ∆FCH that was less than 5 m and greater than 25 m was excluded due to an insufficient amount of pixels. The negative correlation between ∆FCH and ∆LST incidated the enhanced cooling effect with the increase in FCH.

Changes in biophysical components following FCH changes
The underlying biophysical mechanisms behind the observed changes in LST can be better clarified by quantifying the variations in the components of the surface energy balance. Here our analysis was also based on the FCH categories of less than 10 m, 10-20 m, and greater than 20 m. The spatial distributions of mean annual ∆α, ∆R ns , ∆LE, ∆R nl , and ∆H are shown in figures S5-S9. The ∆α values of EF, DF, and MF between different canopy height classes were all negative at annual scale (figures 6(a)-(c)), which confirmed that taller forests had a lower albedo than shorter forests. The ∆α values between different canopy height classes for all forest types were lower in non-growing season than in growing season. Notably, conversion from short MF to medium MF slightly increased the albedo (0.0036 ± 0.0001) during growing season (figure 6(c)). The probable reasons are further analyzed in section 4. For different forest types, the ∆α between different canopy height classes of EF is obviously lower than DF and MF, whether at annual scale, during growing season or during non-growing season.
Contrary to ∆α, the albedo-induced changes in ∆R ns for EF, DF, and MF between different canopy height classes were all positive at annual scale and growing season (figures 6(d)-(f)), which led to a warming effect of taller forests. This warming effect reached its highest in non-growing season due to a more remarkable difference of ∆α in non-growing season than in growing season when FCH converts from shorter to taller. The latitudinal patterns of ∆R ns were more heterogeneous between medium and short forests than the other two classes (figure S6). It is noteworthy that the magnitude of ∆R ns between different canopy height classes for DF and MF was similar except for canopy height transition from short to medium, whether at annual scale, during growing season or during non-growing season. Annually, the magnitude of ∆R ns for EF was almost twice that of DF and MF.
For ∆LE, the values of EF and MF between different canopy height classes were all positive at annual scale (figures 6(g)-(i)), which demonstrated a cooling effect of taller forests. However, a mild warming effect occurred when medium DF was replaced by tall DF (figure 6(h)), which was mainly attributed to the cooling effect of LE in non-growing season counteracted by the stronger warming effect in growing season. Furthermore, the ∆LE between different canopy height classes for all forest types were higher in growing season than in non-growing season, except for DF and MF when FCH transitioned from medium to tall. The lower LE of tall forest than medium forest in growing season led to this result. For different forest types, the ∆LE between different canopy height classes of EF is larger than DF and MF at annual scale and in non-growing season. The similar magnitude of LE between tall and medium MF resulted in the largest ∆LE between medium and short MF compared with EF and DF (figure 6(i)).
The ∆R nl , based on the surface energy balance equation, is contrary to the LST change patterns. Apart from a minor negative value that appeared when converting from medium DF to tall DF during non-growing season, the ∆R nl values of EF, DF, and MF between different canopy height classes were all positive, whether at annual scale, in growing season or non-growing season (figures 6(j)-(l)), leading to a warming effect of taller forests. The consistent change patterns of ∆LST and ∆R nl between tall and short DF during non-growing season was mainly attributed to the impact of emissivity on LW out .
Compared with other surface energy fluxes, the ∆H presented clear spatial and seasonal variability (figures 6(m)-(o)). For example, the magnitude of ∆H was positive during growing season and negative during non-growing season when canopy height class converts from medium to tall. The ∆H between different canopy height classes overall is small (−0.29-−1.49 W m −2 ) due to a large spatial variability, ranging from a maximum value of exceeding 24 W m −2 to a minimum value of lower than −30 W m −2 over CONUS (figure S9). The ∆H showed an opposite overall pattern to ∆LE for EF and DF between different canopy height classes at annual scale. This pattern is due to the surface available energy partitioning between H and LE changes.

Relation of LST change with net radiation, latent heat flux and sensible heat flux changes
To examine whether the changes in biophysical components can explain the ∆LST, the relation between ∆LST and ∆R ns , ∆LE, and ∆H following FCH changes was investigated using multiple regression analysis. Here the relationship between ∆LST and ∆R nl was excluded due to R nl being directly linked to LST. As shown in figure 7, the impact of ∆LE and ∆H on ∆LST was similar for different forest types, whether at annual scale or during growing and non-growing season, whereas the impact of ∆R ns on ∆LST varied depending on forest types and growing periods. For example, the impact of ∆R ns on ∆LST was up to ten times stronger than the ∆LE and ∆H effect on ∆LST when tall MF was compared with short MF during growing season. In contrast, the impact of ∆R ns on ∆LST was similar to the ∆LE and ∆H effect on ∆LST when medium EF compared with short EF during growing season. Obviously, the sign of the regression coefficient for ∆R ns was opposite in growing season and non-growing season. The determination coefficient (R 2 ) of ∆LST with ∆R ns , ∆LE, and ∆H following FCH changes at different timescales is displayed in table 2. Notably, ∆R ns , ∆LE, and ∆H together explained 91% of the variation in ∆LST for EF when canopy height classes transistion from medium to tall during growing season. To sum up, the biophysical components of ∆R ns , ∆LE, and ∆H together explained most of the variation in ∆LST for EF than DF and MF among different canopy height classes, whether at annual scale or during growing and non-growing season. The annual relatively low R 2 between ∆LST and ∆R ns , ∆LE, and ∆H for DF and MF was mainly attributed to the poor performance of these biophysical components in explaining ∆LST during non-growing season.

Discussion
Using satellite observations, we explored the biophysical climate effect of forests with different canopy height classes over CONUS. We further quantified the variations in the biophysical components leading to the observed local surface temperature change. Results showed that at annual scale, the values of ∆LST TS , ∆LST TM , and ∆LST MS for the three forest types (i.e. EF, DF, and MF, respectively) were all negative, indicating the cooling effect of taller forests. Moreover, our results also suggested that the cooling effect of EF is stronger than that of DF and MF, whether for tall, medium, or short FCH classes. The changes in biophysical components indicated that a decrease in the albedo in shorter forests to taller forests resulted in a positive radiative forcing. The surface energy fluxes of net radiation (i.e. R ns and R nl ) and sensible heat flux dominated a warming effect while the latent heat flux regulated a cooling effect when FCH class converts from shorter to taller at annual scale. When FCH transferred from shorter to taller, the albedoinduced warming effect was due to smaller canopy gaps and there being less exposed underlying forest floor in taller forests (Ge et al 2019, Alibakhshi et al 2020). The LE had a cooling effect due to more vigorous ET on the leaf surface in taller forests (Xu et al 2018b). Taller forests are more efficient in transferring water from the deeper soil to the atmosphere through deeper root systems than shorter forests. Yet this is not always the case in this study. For instance, LE led to a mild warming effect when medium DF was replaced by tall DF at annual scale ( figure 6(h)). This may be associated with tall DF suffering from extreme climate (e.g. drought). Previous research demonstrated that the water transportation paths of taller forests are longer than that of shorter forests, leading to higher water demand (Nepstad et al 2007). Therefore, the stomata of tall forests will close to reduce ET and regulate water loss once soil moisture is limited. Nevertheless, tall DF was cooler than medium DF in growing season even though the LE of medium DF was larger than old DF. This was largely attributed to the enhanced cooling effect of H (figure 6(n)). The rough surface of tall forest canopies provides a more efficient turbulent heat exchange with the boundary layer, such that convective cooling satisfies the need for strong evaporative cooling (Teuling et al 2010). Moreover, the albedo of medium MF was larger than that of short MF during growing season (figure 6(c)). The reasons can be illustrated by the following two points. On the one hand, the albedo in the growing season was determined by the combination of canopy near infrared (NIR) reflectance and photosynthetically active radiation (PAR) reflectance (Richardson et al 2013). The amount of NIR in the canopy of taller forests increases with the canopy development during the growing period, as green leaves strongly reflect NIR with an enhancement of multiple scattering from rapid canopy development and LAI (Gates 1965 (Yuan et al 2010, Li et al 2021b. Therefore, we speculated that the medium and young MF in this study was dominated by medium DF and short EF, respectively. The large ∆LE between medium and short MF further supported our speculation (figure 6(i)). The ∆H showed more variable change patterns than other biophysical components despite the climate feedback (warming effect) of ∆H being doubtless at annual scale. In addition to the available energy partition between latent and sensible heat flux, the uncertainty of ground heat flux also contributes to this variability. It has been reported that ground heat flux was linked to soil moisture content, incoming radiation, and H + LE, thus increasing the complexity of the estimation of ground heat flux (Wang and Bou-Zeid 2012, Purdy et al 2016, Duman et al 2021. The stronger cooling effect of EF than DF and MF can be attributed to the following aspects. First, for the same FCH classes of EF, DF, and MF, the proportion of different heights that make up these categories is different. For example, the FCH differences between tall and short EF are larger than that of DF and MF, leading to a higher ∆LST. Second, the climate effect of forests with different canopy heights may be influenced by the background climate or the hydrometeorological state (Pitman et al 2011, Huang et al 2018. The intensified cooling effect of taller EF occurred due to stronger ET in some regions of western United States, where mean annual precipitation is less than 400 mm. DF and MF are mostly located over the eastern United States, where mean annual precipitation is greater than in the western United States. Therefore, the cooling effect of taller DF or MF can be weaker because the precipitation tends to minimize increases of sensible heat flux and reduce the moisture stress limiting the latent heat flux (Zhang et al 2020).
The relationship of ∆LST with biophysical components changes (∆R ns , ∆LE, and ∆H) following FCH changes showed large spatiotemporal variations for different forest types. The poor explanation of ∆LST to these biophysical components changes for DF and MF during non-growing season was likely related to the background climate and environmental conditions (e.g. soil moisture content) and bare soil proportion (Abera et al 2020b). For instance, the pattern of albedo-related ∆R ns was opposite during rainfall extreme periods such as drought and extreme wet events (Abera et al 2020b). The change of CO 2 concentration may affect the water use efficiency of forests and thus have an impact on ET (Hatfield and Dold 2019). These factors affected the partitioning of R ns + R nl into LE and H and hence, in turn, influence the relationship of ∆LST with ∆R ns , ∆LE, and ∆H.
Our results have implications for forest management in tackling climate change. First, it would be crucial for areas where space is limited and it is difficult to expand vegetation cover. An effective way is to consider the climate impact of vegetation height. Second, species of tall EFs with the strongest cooling effect would be preferable for afforestation or forest management to mitigate global warming. Our results are consistent with those studies in that tall vegetation is more efficient in decreasing local surface temperature (Ren et al 2018, Yu et al 2018, Li et al 2021a. Meanwhile, our study has gone a step further than previous studies in two aspects. On the one hand, compared with most studies focusing on the biophysical effect of land cover change (e.g. afforestation, reforestation, and deforestation) on LST, research concerning forest canopy structure changes on LST is reported less. This study quantified the biophysical climate impact of forest canopy structure changes on LST from the perspective of FCH, which can be considered to complement to related studies on biophysical climate effect. On the other hand, compared with studies aiming at urban scale, this study fills in the gaps on the biophysical effects of forests under natural conditions at the national scale. Besides, the underlying biophysical mechanism of FCH changes in influencing LST was clarified from the perspective of surface energy balance. Nonetheless, it should be noted that our study also has some limitations. Previous studies exploring the biophysical climate impact of land cover changes have proposed several energy balance-based decomposition methods such as the decomposed temperature metric and intrinsic biophysical mechanism to quantify the contributions of biophysical components on LST (Lee et al 2011, Luyssaert et al 2014, Abera et al 2020a, Zhang et al 2020. These studies clearly illustrated the changes in energy fluxes (shortwave, sensible and latent heat fluxes), which were consistent with our study. However, this study was incapable of applying the above-mentioned physically-based attribution approach to elucidate the relative contributions of biophysical components to the FCH changesdriven local LST change. This was due to a lack of incoming longwave radiation and sensible heat flux products with high spatial resolution. Furthermore, the aggregation of land cover data and resampling of shortwave radiation and emissivity data might introduce uncertainty into our results. Last but not least, the approach this study used is space for time, which means FCH classes are constant over a period of time.
In the future, with the development of the annual remote sensing dataset for long time series, it will be possible to investigate forests' biophysical climate effect together with the 'space and time' strategy.

Conclusions
Using satellite observations from between 2003 and 2005, we investigated the biophysical effect of forests with different canopy height classes on local LST and illustrated the biophysical processes controlling LST change over CONUS. Our analysis was based on the space-for-time approach in combination with the energy balance equation. Moreover, multiple regression analysis was performed to examine whether the changes in biophysical components can explain the ∆LST. Results showed that FCH has a significant impact on LST. For diverse forest types (EF, DF, and MF), taller forests are cooler than shorter forests at the annual scale. This annual cooling effect is mainly attributed to enhanced ET outweighing albedo and sensible heat flux changes when FCH classes convert from shorter to taller. Furthermore, the effect of FCH on LST is different among forest types. The cooling effect of EF is stronger than that of DF and MF, whether for tall, medium, or short FCH classes. This is perhaps due to DF and MF being mostly distributed over the eastern United States where the wet conditions tend to prevent moisture limitations to latent heat flux. The multiple regression analysis indicated that the biophysical components of ∆R ns , ∆LE, and ∆H explain the variation of ∆LST better in growing season than in non-growing season. The poor performance during non-growing season is mainly affected by background climate and environmental conditions. Our results highlight the necessity of considering FCH when studying forests' biophysical climate effects. It also provides valuable theoretical guidance for forest management regarding tackling climate change.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).