Solar dimming above temperate forests and its impact on local climate

Forest inventory data for Trentino was provided by the 'Servizio Foreste e Fauna' of the province and it comprised net surface of forest area and growing stock for the period 1977–2015.

Global radiation (G) data measured at high-elevation stations from Trentino (in the montane to subalpine zone) were obtained from www.meteotrentino.it/ and from the Fondazione Edmund Mach. They are briefly described in the online supplementary table 1 available at stacks.iop.org/ERL/13/064014/mmedia and their geographical distribution is shown in supplementary figure 1. Mean annual values were calculated for all the stations and the trend in irradiance was determined by means of GLM (General Linear Model) analysis. Years with less than 85% of available daily data were not included in the calculation. The monthly trends over the entire available period were also computed using only months with less than 15% missing data (5 days missing).

FLUXNET site pairs consisting of a forest and a neighbouring cropland or grassland were selected for analysing differences in incoming radiation and air temperature between different ecosystems. The site pairs approach allows a direct evaluation of the differences between different land covers (Vanden Broucke et al 2015). The criteria for the site selection were:

• environmental and geographical location: the pair must be composed of one site located in a forest and the other one in agricultural land/grassland; the full ensemble of the pairs needs to be located in the mid-latitudes, where the net effect of afforestation is not precisely understood; the distance between the sites must not exceed the scale of mesoscale air masses circulation;
• data availability: common data availability for at least 5 years;
• areas located in flat terrain were preferred to avoid additional relief effect on the radiation balance.

The pairs which satisfied the above criteria are briefly described in table 1 and their location is shown in figure 1.

Zoom In Zoom Out Reset image size

The level 2 product data (half-hourly data, quality checked by the PI of the site but not gap filled) for the indicated periods (table 1) were obtained from www.europe-fluxdata.eu/. In the case of the TK site pair, a third site was additionally selected (the grassland Grillenburg). The comparison between the measurements above the grassland Grillenburg (58 ha) and above the surrounding Tharandt forest (5800 ha), should help to illustrate the extent of forest's atmosphere feedbacks on adjacent ecosystems.

In a pre-processing step, the influence of missing data was tested for each site separately by simulating data gaps in complete measuring series (the procedure is described in detail in the supplementary material). The aim was to minimize the effect of data gaps (due to temporary sensor removal for calibration, malfunction etc) on the analyses.

The sites were analysed both, pairwise and aggregated to land use groups, to assess the existence and the significance of differences in incoming global solar radiation (G, 0.3–4.5 µm). Since each of the investigated forests was composed of different tree species, analysing the pairs individually provides an idea of the tree species influence on the mentioned parameters. On the other hand, grouping the sites by land use helps to identify patterns of the land cover influence.

A direct difference (forest minus cropland/grassland) between the half hourly measuring values was calculated for incoming global radiation (G) and air temperature (TA) for each site pair. Frequency distributions were calculated for the resulted differences using narrow difference classes (class width of 10 Wm⁻² and 0.1 °C respectively). For each resulting frequency class (negative or positive) a mean difference in global radiation (forest minus non-forest) was calculated as:

$$\begin{equation} {{\rm{\Delta }}_{\rm{c}}}{\rm{ = m \times n}}/{\rm{N}} \end{equation}$$

where m is the centre of each frequency class, n is the number of differences in a frequency class and N is the total number of considered half hourly values.

The aim of this approach was to evaluate the magnitude of the net differences in irradiance over various classes of differences, or in other words to understand when the differences in irradiance were larger and when smaller and when they were positive or negative. Therefore a net difference was further computed for the absolute values of each class (i.e 0–10, 11–20, 21–30...) as

$$\begin{equation} {\rm \Delta}_{\rm{nc}} = {\rm \Delta}_{{\rm c}(+)} + {\rm \Delta}_{{\rm c}(-)} \end{equation}$$

where ∆_c(+) and ∆_c(−) are the values obtained for the positive difference classes and for the corresponding negative difference classes respectively.

This procedure was applied on the entire study period, on an annual and monthly basis. The net effect for a specific time step was calculated as:

Δ_tot = ΣΔ_(tot) = the net difference for the entire studied period

Δ_a = ΣΔ_nc(a) = the net difference for the entire studied period

Δ_m = ΣΔ_nc(m) = the net difference on monthly level.

A graphical example of the approach is shown in supplementary figure 3. In the following text ∆G_a, ∆G_m, ∆G_tot and ∆TA_a ∆TA_m, ∆TA_tot stand for annual, monthly and total differences of global radiation and air temperatures, respectively.

The annual values were used to correlate ∆G_a (independent variable) to the annual differences in air temperature (∆TA_a, dependent variable) for each site. The aim of this analysis was to obtain the individual trend for each pair but also to build a common trend over all the sites. To compensate for possible systematic bias in air temperature due to differences in site properties between the pairs, e.g. elevation, the intercept of each individual regression function (i.e. for each pair) was removed, therefore assuming that a change of 0 W m⁻² in ∆G_a corresponds to a change in air temperature (∆TA_a) of 0 °C at all sites. This procedure also set the common basis to analyse all site pairs together.

For other parameters like sensible heat flux (H), latent heat flux (LE) and albedo, mean annual courses were assembled from daily means.

In the specific case of the site pair Hainich-Gebesee (where the sites are separated by 270 m of elevation) additional analysis was made to account for the possible influence of elevation difference on the differences in irradiance. Air masses coefficients—defined as the direct optical path length of solar irradiance through the earth atmosphere—were computed for the elevation of the two stations (according to Kasten and Young 1989). Furthermore, a theoretical difference in clear sky irradiance (considering a 10% proportion of diffuse radiation) was calculated using the mean coefficients for polluted air according to Meinel and Meinel (1976).

For each site pair, an afforestation effect was simulated adapting the modelling approach used by Lee et al (2011), which simulates a deforestation effect on the surface air temperature using site pairs consisting of forests (FLUXNET forest towers) and surface weather stations located in grasslands as a proxy for open land. The main assumptions made therein are: the two sites forming a pair receive the same amount of incoming shortwave (SW_in) and incoming longwave radiation (LW_in), air is sufficiently mixed at blending height h so that T_air is identical. Minor influences produced by differences in emissivity or soil heat flux are not considered.

The approach used in this paper attempts to assess the inverse effect (afforestation) by similar means: i.e. site pairs approach, making similar main assumptions, but accounting also for the observed change in incoming radiation.

This implied the calculation of a total forcing (∆TF) as a cumulative effect of the difference in incoming radiation (∆G) and the intrinsic biophysical mechanisms involved by land cover changes using the equation:

$$\begin{equation} \begin{array}{l@{}l@{}l@{}} {\rm{\Delta TF}}\left[ {{\rm{W}}{{\rm{m}}^{{\rm{ - 2}}}}} \right] =\\ {\rm{ \Delta G + \Delta R}}{{\rm{F}}_{\left( {{\rm{albedo}}} \right)}}{\rm{ + \Delta E}}{{\rm{R}}_{\left( {{\rm{roughness}}} \right)}}{\rm{ + \Delta E}}{{\rm{R}}_{\left( {{\rm{Bowen}}} \right)}} \end{array} \end{equation} \tag{ 1 }$$

where

∆G = the difference in incoming global radiation (0.3–4.5 µm) forest-grassland/cropland

∆RF_(albedo) = the radiative forcing due to albedo changes

∆ER_(roughness) = energy redistribution caused by changes in roughness

∆ER_(Bowen) = energy redistribution caused by changes in Bowen ration

ΔRF_(albedo) = ΔS/1 + f where ∆S is the difference in absorptivity between the forest and the grassland

f is an energy redistribution factor (computed as explained in the supplementary material)

ΔER_(roughness) = Rn × Δf₁/(1 + f)² where Rn is the net radiation, ∆f₁ is the change in energy redistribution due to changes in roughness

ΔER_(Bowen) = Rn × Δf₂/(1 + f)² where ∆f₂ is the change in energy redistribution due to changes in Bowen ratio.

∆f₁, ∆f₂ (the changes in energy redistribution factor f) and f were computed according to the methodology of Lee et al (2011). A more detailed explanation of the equations that were used is presented in the supplementary material.

Daily mean values were used for all computations. Furthermore a common correlation ∆TF versus ∆TA_a was build using the same procedure as for the correlation between ∆G_a and ∆TA_a.

Above canopy data for the forests were only available for mature forest ecosystems and not for other development stages. Thus an additional assumption of our model is that the grass/cropland is replaced by a mature forest, ruling out the consideration of additional effects specific for earlier development stages, like for example elevated transpiration (Moore et al 2004), higher stomatal conductance, slightly higher albedo etc. Nonetheless, the calculated total forcing should exhibit the general implications of converting a grassland/cropland to a forest.

The net forest area in Trentino increased by approximately 10% and by more than 50% in growing stock since 1977. During the last and a half decade the increase accounted for about 2% in net forest cover and approximately 20% in growing stock (figure 2(a)). Global radiation (G) measured at montane to sub-alpine zone stations in Trentino (supplementary figure 1) consistently decreased over the period 2000–2015 (figure 2(b)), thus confirming previous observations made for a shorter period of time by Tudoroiu et al (2016). Solar dimming was more pronounced in summer, particularly in June (supplementary figure 5).

Zoom In Zoom Out Reset image size

Global incoming radiation was generally less over forests than over the other land cover types (cropland and grassland) (figure 3(a), supplementary figures 7–9). Annual means of ∆G_a (computed as forest minus grassland/cropland) were mostly negative for all the sites (figure 3(b)). Mean values over the studied period are shown in table 2. The analysis of monthly data revealed that the HG and TK site pairs were different from the LC (figure 3(c)) site pair. The largest monthly differences in incoming global radiation (∆G_m) were observed in winter time for HG and TK, and in spring for LC (figure 3(c), supplementary figures 10–12). When all the site pairs were merged (figure 4), the differences in incoming global radiation ∆G_a were correlated to the corresponding observed differences in air temperature (∆TA_a, p = 0.0002). A variation of one W m⁻² of incoming radiation corresponded to a change of 0.030 ± 0.02 °C in air temperature. Only minor and non-significant mean differences in G_a were observed for the TG (Tharandt/Grillenburg) site pair (supplementary figure 6). The latent heat flux (LE) was on average slightly higher above the forest for all the site pairs (figure 5(a), table 2). The sensible heat flux (H) was on average always higher above the forest (figure 5(b), table 2). Surface albedo was lower over the forest (figure 5(c), table 2). The difference in air masses coefficients caused by the difference in elevation at the HG site pair was equivalent to 35 Wm⁻² for clear sky irradiance.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The contribution of each term of equation 1 to the total forcing (∆TF) is shown in figure 6. ∆G_a is negative for all the site pairs (table 2). Albedo (∆RF_(albedo)), and Bowen ratio (∆ER_(Bowen)) forcings were always positive. The forcing caused by differences in surface roughness (∆ER_(roughness)) was always negative. When the differences in the observed radiation are also considered, the mean ∆TF was negative for HG and TK but slightly positive for LC (figure 6, table 2). A significant correlation was found between ∆TA_a and ∆TF (p < 0.00001). Air temperature changed by 0.032 ± 0.01 °C per Wm⁻² of TF (figure 7). ∆TF was instead largely positive for the TG site pair, as in this case the difference in radiation ∆G_a was negligible (table 2).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

The effects of afforestation or deforestation on local climate have been often assessed by comparing forests and neighbouring non-forest sites (i.e. Teuling et al 2010, Lee et al 2011, Baldocchi and Ma 2013, Zhang et al 2014, Luyssaert et al 2014, Vanden Broucke et al 2015). Paired forests and non-forests sites are hardly found in mountain areas where, for instance, the radiosity effects of the slopes are likely to play a role. This justifies the selection of paired sites in regions where orographic effects are negligible. The Central European site pairs that are analysed here strongly supported the view that forests may cause a consistent and significant dimming (figure 3). Since the two sites forming a pair are separated by a certain spatial distance, the chance that they were subject to different synoptic conditions cannot be totally excluded. Differences in half-hourly radiation can be largely negative or positive when the sky at one of the two sites is overcast while it is clear at the other one. Binning the half hourly radiation differences into classes showed that, the frequency of those conditions was mostly equivalent in all site pairs, while incoming radiation was significantly different mostly when the differences in irradiance between the two sites were 0–350 Wm⁻² (figure 3, supplementary figures 7–9). The presence of thicker clouds over forest in the cloudy episodes, the occurrence of more haze over forest in the sunny episodes or more shallow clouds over forests might explain such differences.

The elevation difference between the forest and the cropland at the HG site pair (approximately 270 m, table 1) may have caused additional effects. Such a difference in elevation corresponded to 0.03–0.6 relative air masses for high and low sun heights, respectively. This would imply that, simply as an effect of elevation difference, the irradiance over the Hainich forest should be potentially 35 W m⁻² higher. However, the comparison of the observed differences (forest minus cropland) for clear sky days (sun heights greater than 10 degrees) with the corresponding calculated theoretical differences, showed that the observed net difference reached on average only 37% of the expected 35 W m⁻². This confirms the diminishing effect of the forest on the incoming radiation. It cannot be totally excluded that over the hills, where the Hainich forest is located, the overall cloud cover percentage was higher. On one hand, clouds passing through can persist due to the orographical barrier, but on the other hand, it can be exactly the postulated land use effect.

The examination of the 'special' additional site pair (TG, Tharandt/Grillenburg) reinforced the results obtained for the other three site pairs. Here, the advection of 'forest air' is likely prevalent also above the grassland area of Grillenburg which is entirely nested into the Tharandt forest. In this case, the mean difference in irradiance over the studied period for the two sites was nearly zero (supplementary figure 6). This indicates that the extent of the grassland was too small to have a feedback on the local irradiance climate.

In this line of arguments the forest expansion in Trentino, reflected in both, increase in area and growing stock and which went on for over three decades (figure 2(a)) can well have triggered the observed dimming, even if the two effects did not occur simultaneously.

The differences in incoming global radiation between forest and non-forest land were larger during winter for the site pairs HG and TK (figure 3(c), supplementary figures 10–12). This result is apparently surprising, since studies on secondary plant metabolism and on the temperature dependence of BVOCs emission by plants reported that their emissions are generally higher in the warm season (Guenther et al 1995, 2006). However, it has also been shown that leaf litter can be a significant emitter of monoterpenes (Faiola et al 2014) and that needle litter in the first stages of decomposition (up to 165 days) also emits monoterpenes at rates comparable to those emitted by attached needles (Isidorov et al 2010). Accordingly, emissions of BVOCs from forests do also occur during winter. Nevertheless, other confounding effects may come into play during summer: non-forest areas in the proximity of Gebesee and Klingenberg are croplands and it is well known that crop management, such as tillage and intensive mechanical soil cultivation, occurs mainly in spring and summer (i.e Funk et al 2008). Tillage operations inevitably release particulate matter (PM) (dust and bioparticles) which can be 5 times higher during summer months compared to spring (Funk et al 2008). Those anthropogenic particulate emissions are therefore likely to offset part of the BVOC-driven solar dimming effect of the forests, by attenuating incoming radiation where agricultural areas and roads prevail. Such an effect was not recorded at the site pair LC, where the land cover in the proximity of the Cabauw station (non-forest site) is mainly grassland, which is generally less intensively managed than croplands. In this case, the largest differences in irradiance between forest and non-forest land were observed during late spring and early summer. In this period of the year, the growth rates and monoterpene emissions of Scots pine (the dominant tree species of the Loobos forest) are expected to be the highest (Holzke et al 2006) and dust emission from less intensively managed land can hardly offset the forest-driven solar dimming effect. It must be outlined, however, that the Cabauw station is closer to the sea (60 km) than Loobos (figure 1). Wintertime storms often transport marine aerosol inland and light attenuation effects are often larger when the distance from the coastline is smaller. Therefore it cannot be excluded that the larger values of ∆G_m during summer for the site pair LC may be also influenced by the occurrence of periodic gradients of marine aerosols concentrations.

In Trentino we observe a maximum reduction in global radiation in summer (June, supplementary figure 5) when BVOC emissions are expected to be the highest (Guenther et al 1995, 2006). Here, the lack of crop management activities and different phenology of high elevation forests can explain the different dimming seasonality compared to the Central European site pairs.

A lot of attention was paid to the question whether forests cool or warm the air or, in other terms, if forests mitigate or exacerbate global warming at the local and regional scales. There is a general consensus that afforestation exacerbates warming in snow-rich areas, as persisting snow cover over grasslands and other non-forest areas cause prolonged high-albedo conditions. On the contrary, the snow cover rapidly disappears in forests when the accumulated snow falls from the tree branches soon after a snow event (Schwaab et al 2015). Bonan (2008), among others, argued that afforestation may lead to substantial warming in Northern latitudes, but also to substantial cooling in the Tropics and uncertain effects in mid latitudes. Alkama and Cescatti (2016) showed instead that a decrease in forest area caused an increase in near-surface air temperature in all the regions of the world, especially in terms of maximum temperatures. The observation of an overall decrease in global incoming radiation above forests, as considered in this study, would suggest in a first instance that a decrease in incoming energy could somewhat translate into an overall cooling effect. Forests have a lower albedo and a higher surface roughness compared to other land covers (Baldocchi and Ma 2013). Lower albedo is the consequence of the absorption of more sunlight by the surface and such additional absorbed energy must be somewhat redistributed into sensible, latent and soil heat flux. On the other hand enhanced surface roughness facilitates turbulent heat dissipation either as sensible or latent heat flux (Rotenberg and Yakir 2010, Lee et al 2011). In the specific case of the three site pairs, the albedo was always lower in case of the forest and this translated on average into a higher longwave emission. The higher surface roughness of the forests resulted in generally higher sensible heat fluxes (figure 5).

These results are confirmed by the modelling approach employed here. The darker surface and the higher Bowen ratio of the forests led to a positive ∆RF_(albedo) and ∆ER_(Bowen) (figure 6, table 2). On the other hand, forests have a lower aerodynamic resistance compared to other ecosystems leading to a negative ∆ER_(roughness). In the case of the site pairs HG and TK the dimming induced by the forests prevailed over the other terms causing a negative ∆TF (cooling effect). At the LC site, the ∆ER_(Bowen) is higher compared to those at the HG and TK, showing the differences in latent and sensible heat between the grasslands (Cabauw) and croplands (Gebesee and Klingenberg). In this case, the dimming effect was balanced by the positive ∆RF_(albedo) and ∆ER_(Bowen) forcing and caused a small positive ∆TF (figure 6, table 2). Another interesting modelling result is that broadleaf forests, such as the European beech forest at Hainich, more likely produce a negative forcing than coniferous forests (Tharandt and Loobos forests), mainly as a result of a higher albedo and lower Bowen ratio.

The significant correlation found between ∆TF and the corresponding differences calculated for air temperature (∆T_a, figure 7) shows that, a one unit change in ∆TF (1 Wm⁻²) causes a change of 0.032 ± 0.01 °C in near-surface air temperature. This effect is comparable to what was found in a regression analysis between ∆G and ∆TA_a (0.030 ± 0.02 °C). An indirect confirmation of the strong effect of forests on solar radiation is the positive ∆TF recorded at the TG site pair. The air quality over the small grassland patch embedded in a large forest complex is presumably the same than over the forest and thus almost no dimming effect was observed. The sum of the other biophysical mechanisms results in a positive ∆TF.

In a global scenario, where no land cover change is considered, the radiative forcing caused by changes in global radiation is assumed to be in the order of 0.2 °C/Wm⁻² (Turnock et al 2016). Bearing this in mind we can assume that for the analysed site pairs, the modelled biogeophysical changes induced by afforestation (changes in albedo, roughness and Bowen ratio) reduced the predicted global radiation driven radiative forcing by approximately one order of magnitude. Nonetheless, based on this line of arguments, it may be concluded that modelled forest-driven solar dimming can lead to a small but detectable cooling of the air over the forest.