Biophysical forcings of land-use changes from potential

Ecological Monographs, 84(2), 2014, pp. 329–353 Ó 2014 by the Ecological Society of America

Biophysical forcings of land-use changes from potential forestry activities in North America
KAIGUANG ZHAO1,2,4 AND ROBERT B. JACKSON1,3
1Division of Earth and Ocean Sciences and Center on Global Change, Nicholas School of the Environment, Duke University, Durham, North Carolina 27708 USA
2School of Environment and Natural Resources, Ohio Agricultural Research and Development Center, The Ohio State University, Wooster, Ohio 44691 USA
3School of Earth Sciences, Woods Institute for the Environment, and Precourt Institute for Energy, Stanford University, Stanford, California 94305 USA
Abstract. Land-use changes through forestry and other activities alter not just carbon storage, but biophysical properties, including albedo, surface roughness, and canopy conductance, all of which affect temperature. This study assessed the biophysical forcings and climatic impact of vegetation replacement across North America by comparing satellitederived albedo, land surface temperature (LST), and evapotranspiration (ET) between adjacent vegetation types. We calculated radiative forcings (RF) for potential local conversions from croplands (CRO) or grasslands (GRA) to evergreen needleleaf (ENF) or deciduous broadleaf (DBF) forests. Forests generally had lower albedo than adjacent grasslands or croplands, particularly in locations with snow. They also had warmer nighttime LST, cooler daily and daytime LST in warm seasons, and smaller daily LST ranges. Darker forest surfaces induced positive RFs, dampening the cooling effect of carbon sequestration. The mean (6SD) albedo-induced RFs for each land conversion were equivalent to carbon emissions of 2.2 6 0.7 kg C/m2 (GRA–ENF), 2.0 6 0.6 kg C/m2 (CRO–ENF), 0.90 6 0.50 kg C/m2 (CRO–DBF), and 0.73 6 0.22 kg C/m2 (GRA–DBF), suggesting that, given the same carbon sequestration potential, a larger net cooling (integrated globally) is expected for planting DBF than ENF. Both changes in LST and ET induce longwave RFs that sometimes had values comparable to or even larger than albedo-induced shortwave RFs. Sensible heat flux, on average, increased when replacing CRO with ENF, but decreased for conversions to DBF, suggesting that DBF tends to cool near-surface air locally while ENF tends to warm it. This local temperature effect showed some seasonal variation and spatial dependence, but did not differ strongly by latitude. Overall, our results show that a carbon-centric accounting is, in many cases, insufficient for climate mitigation policies. Where afforestation or reforestation occurs, however, deciduous broadleaf trees are likely to produce stronger cooling benefits than evergreen needleleaf trees provide.
Key words: albedo effect; biophysical forcing; carbon accounting; carbon sequestration; climate regulation; ecosystem services; forestry; land-use change; radiative forcing.

INTRODUCTION
Accompanying the need to combat global warming is an increasing interest in how ecosystems regulate climate (e.g., Heimann and Reichstein 2008). Along with traditional goods and services, such as biodiversity conservation and watershed protection, the climatic benefits of ecosystems are generally assessed from a carbon-centric perspective (McAlpine et al. 2010). Alterations to ecosystems can indeed change carbon sinks or sources that dampen or accelerate global warming. Since 1850, for instance, land-use change has released ;150 billion metric tons of carbon, accounting for 35% of anthropogenic CO2 emissions (Houghton
Manuscript received 2 October 2012; revised 6 August 2013; accepted 11 September 2013; final version received 19 October 2013. Corresponding Editor: A. O. Finley.
4 E-mail: [email protected]

2003). Safeguarding and enlarging terrestrial carbon pools are thus key strategies to mitigate climate change, typically through forestry practices such as reforestation, afforestation, avoided deforestation, and forest management (e.g., Jackson and Baker 2010, McKinley et al. 2011).
Land alterations by forestry and other activities modify not only carbon stocks, but also energy partitioning, water cycling, and atmospheric composition (Fig. 1). These changes occur through altered biophysical characteristics, including albedo, surface roughness, sensible and latent heat fluxes, canopy conductance, soil moisture, surface temperature, emissivity, leaf area, and rooting depth (Kueppers et al. 2007, Anderson et al. 2011, Jayawickreme et al. 2011). For instance, forested surfaces often have lower albedo and more uneven canopies compared to other vegetation, absorbing more sunlight and facilitating the mixing

329

330

KAIGUANG ZHAO AND ROBERT B. JACKSON

Ecological Monographs

Vol. 84, No. 2

of air (Betts 2000). Locally, these climate forcings affect temperature more than the CO2 reduction does. At regional and global scales, biophysical forcings can either amplify or diminish the cooling benefit of carbon uptake. Because of such interactions, researchers have recently recommended that climate policies for crediting forestry projects should go beyond a carbon-centric accounting to include biophysical effects (Jackson et al. 2008, Montenegro et al. 2009, McAlpine et al. 2010).
Although considerations of both biophysical and biogeochemical mechanisms are undoubtedly important in formulating policies to optimize climate benefits of forestry or land management activities, the science for such integration is still evolving (e.g., West et al. 2011). One unresolved issue is how best to capture the spectrum of climate forcings for biophysical and biogeochemical processes that tend to occur at vastly different spatial and temporal scales (Bonan 2008). For instance, the climate effects of carbon sequestration are global and long lasting and are quantified primarily in terms of radiative influences on global mean temperatures. In contrast, biophysical impacts are dominantly local or regional; they occur with altered lands and diminish if the lands revert. Biophysical changes also exert both radiative and non-radiative influences, modifying air temperatures and hydrologic cycles. These disparities in mechanisms raise issues as to how to combine biophysical and biogeochemical regulations into policy measures for climate change mitigation.
Comparisons of carbon sequestration and biophysics for climate regulation by ecosystems are typically assessed in terms of radiative forcing (RF), defined as the perturbation to the radiation balance of the climate system (Betts 2000, Rotenberg and Yakir 2010). With tree planting, reduced albedo and carbon uptake typically cause a positive shortwave (warming) and a negative longwave (cooling) RF, respectively. The opposite is often true when clearing forests for other land uses. Agricultural land use during the past 300 years is estimated to have led to a global RF of À0.15 W/ m2 and a cooling of À0.098C due to biophysical effects (Matthews et al. 2003). Additionally, climatic consequences of natural disturbances to forests, such as fire, insect infestation, windfall, and drought, have been examined with RF or its equivalent carbon metrics, incorporating the effects of both carbon release and albedo change (O’Halloran et al. 2011).
Vegetation replacement also alters the exchange of energy and matter between ecosystems and the atmosphere, particularly through the re-partitioning of sensible and latent heat (Juang et al. 2007). These nonradiative forcings modify the boundary layer and influence local climate (West et al. 2011). Increased sensible heat flux warms the near-surface air and the mixed layer directly. Increased evapotranspiration (ET) of trees not only moistens the air, but can also offset the extra solar absorption incurred by lower albedo (Nosetto et al. 2011), thus tending to cool the surface locally

and sometimes the near-surface air. This evaporative cooling varies with season and place, being most pronounced in tropical forests (Anderson et al. 2011). Moreover, alterations in ET mediate land-air interactions through potential changes in lapse rate, longwave RF, and cloudiness.
Observations and earth-system models are both powerful tools for examining the climatic footprint of land-use change (Bonan 2008). For instance, climate simulations indicated a global cooling effect from replacing short vegetation with forest, attributable mainly to the enhanced ET that fostered low-level cloudiness and attenuated sunlight (Ban-Weiss et al. 2011). Paired model simulations have also suggested that deforestation should be avoided in the tropics and reforestation discouraged at high latitudes to harness climatic benefits of trees, although the latter result is controversial (Randerson et al. 2006, Bala et al. 2007). Meanwhile, differences in climate model structure and parameterization sometimes generate conflicting results (Jackson et al. 2005, Diffenbaugh 2009). In particular, uncertainties exist as to whether the biophysical effects of reforestation in temperate zones will strengthen or weaken the cooling from carbon sequestration (Betts 2000, Jackson et al. 2008, Montenegro et al. 2009).
Despite growing recognition of the biophysical regulation of climate by ecosystems, quantifying their effects is challenging for academic researchers, let alone for resource managers and policy makers (McAlpine et al. 2010, West et al. 2011). Existing efforts to quantify biophysical regulations have typically considered albedo but neglected other important biophysical forcings. For instance, altered ET and land surface temperature (LST) induce longwave RFs that can sometimes be comparable to the albedo-induced shortwave RF (Swann et al. 2010). Improved assessments are needed for biophysical forcings of land-use changes and their policy implications.
We combined remotely sensed observations and climate model outputs to examine the biophysical forcings and climatic impacts of potential land-use/ land-cover changes across North America. We emphasized transitions from non-forest to forest vegetation relevant to climate mitigation policies, specifically cropland (CRO) and grassland (GRA) conversions to evergreen needleleaf forest (ENF) and deciduous broadleaf forest (DBF). We examined surface variables important for temperature and energy balance, including albedo, LST, and ET (Table 1). We evaluated the magnitudes and directions of differences in these biophysical variables between adjacent sites of contrasting vegetation across North America between 208–608 N, using paired comparisons to assess the changes in surface biophysics associated with land conversions. Observed differences were then used to (1) calculate shortwave and longwave RFs or equivalent carbon emissions, (2) infer the redistribution of surface energy for conversions from GRA or CRO to ENL or DBF,

May 2014

LAND USE, BIOPHYSICS, AND FORESTRY

331

FIG. 1. Besides carbon, biophysics matters in assessing climate benefits of forestry projects: Forests and non-forest vegetation have contrasting biophysical properties, resulting in differing land–air interactions. Compared to non-forest lands, forests typically (1) have lower albedo and absorb more solar energy; (2) often have higher surface roughness, facilitating the exchange of water and heat between surfaces and the air; (3) are often cooler, emitting less thermal radiation; and (4) have higher leaf areas and deeper roots, likely increasing evapotranspiration. Larger latent heat fluxes and smaller sensible heat fluxes over forests can decrease the lifting condensation level (cloud base height), thus lowering cloud height and increasing the chance for cloud formation. The relative magnitudes of surface energy fluxes for the four vegetation types studied here (grasslands [GRA], deciduous forests [DBF], croplands [CRO], and evergreen forests [ENF], as depicted clockwise from the top in the graph) are indicated by the sizes of arrows. These biophysical differences highlight that reforestation and afforestation impact climate via biophysical pathways in addition to carbon sequestration.

and (3) assess potential impacts on near-surface temperature. A primary goal of our work is to foster a more complete accounting of climate regulation for ecosystem and land-use management and policy mitigation.
METHODS
Data
This study focused on the vegetated lands of North America between 208–608 N including the conterminous USA (Appendix: Fig. A1). A range of surface and atmospheric variables derived from remote-sensing observations was compiled from three geoportals: the ModerateResolution Imaging Spectroradiometer (MODIS; data available online)5 land surface products, the Clouds and the
5 https://lpdaac.usgs.gov/data_access

Earth’s Radiant Energy System (CERES; data available online)6 data archive, and the Advanced Microwave Scanning Radiometer-Earth Observing System (AMSEE; data available online).7 The MODIS data we used comprised yearly land cover (MOD12Q1; Friedl et al. 2002), eight-day 500-m bidirectional reflectance distribution function (BRDF)/Albedo (MCD43A1 and A2 Collection 5; Schaaf et al. 2002), eight-day 1-km daytime and nighttime LSTs (MOD11A2 Collection 5; Wan et al. 2004), and eight-day 1-km ET (MOD16A2; Mu et al. 2011). The MODIS land-cover data included both 500-m Collection-5 maps for years 2000–2008 and a 1-km Collection-4 map for the year 2001 (Friedl et al. 2002), with the latter being used as a baseline vegetation map for
6 https://eosweb.larc.nasa.gov 7 http://nsidc.org/data/ae_dysno

332

KAIGUANG ZHAO AND ROBERT B. JACKSON

Ecological Monographs

Vol. 84, No. 2

TABLE 1. Major concepts and terms pertinent to the quantification of biophysical forcings of land-use change from forestry activities, with common units supplied whenever applicable.

Concepts and terms Climate regulation Climate forcing Radiative forcing Climate sensitivity Climate efficacy Non-radiative forcing Biophysical forcing
Albedo Emissivity Sensible heat flux Latent heat flux
Land surface temperature Near-surface air temperature C sequestration potential
Carbon-emission equivalent
Net carbon drawdown
Greenhouse gas value

Common units W/m2 W/m2 K/[W/m2] unitless W/m2 W/m2
unitless unitless W/m2 W/m2
K or 8C K or 8C kg C/m2 or
kg CÁmÀ2ÁyrÀ1 kg C/m2 or
kg CÁmÀ2ÁyrÀ1
kg C/m2 or kg CÁmÀ2ÁyrÀ1
kg C/m2 or kg CÁmÀ2ÁyrÀ1

Explanation
Ecosystems offer regulating services by influencing climate via both biogeochemical and biophysical pathways.
An energy imbalance imposed on the climate system either naturally or by human activities, such as C emissions arising from altered ecosystem structure.
The change in radiative energy flux resulting from climate forcing agents such as a CO2 increase or albedo change. Positive radiative forcings, either longwave or shortwave, increase global mean temperature.
A measure of how responsive the climate system is to the radiative forcing of a forcing agent. It is often quantified as the increase in global mean air temperature given a unit of radiative forcing
The global temperature response per unit radiative forcing of an agent relative to that of CO2. It is defined as the ratio of climate sensitivity between a forcing agent and CO2 change.
An energy imbalance that does not directly involve radiation, such as the increase in evapotranspiration due to irrigation.
The imbalance of energy fluxes resulting directly or indirectly from changes in biophysics, including albedo, emissivity, sensible and latent heat, and surface roughness. These biophysical changes can be caused by both natural and anthropogenic processes such as land conversion, ecosystem disturbances, and ecosystem management.
Reflectivity of sunlight by land surfaces, as contributed from both soils and vegetation. Conversions of croplands or grasslands to forests often reduce surface albedo and induce positive shortwave radiative forcings, diminishing the cooling benefit of forest carbon sequestration.
The relative ability of land surfaces to emit thermal radiation. Forests often have slightly higher emissivity than do croplands or grasslands. The biophysical forcing of altered emissivity from land-use change is typically much smaller than that of altered albedo.
The flux of heat between land and the air via conduction and convection. Sensible heat directly warms the air. Altered sensible heat flux due to vegetation shift is a direct warming or cooling effect on local climate.
The flux of heat between land and the air via evapotranspiration or condensation. Latent heat doesn’t directly warm the air. Altered latent heat flux due to vegetation shift is a nonlocal biophysical forcing, which modifies surface energy balance, the hydrological cycle, atmospheric water vapor, and cloud formation.
The temperature of the composites of vegetation and soils over vegetated surfaces, which can be defined either radiometrically, thermodynamically, or aerodynamically. Changes in vegetation structure affect surface energy partitioning and thus strongly affect land surface temperature.
The temperature of the air two meters above a vegetation-specific displacement height for a vegetated surface. This is the temperature metric used here to directly evaluate the temperature effect of land-use change.
The amount of carbon potentially drawn from the air for a given forestry project due to biological carbon sequestration. Its exact value is difficult to estimate and in this study is considered simply as the difference in steady-state total carbon stock between the forest and the replaced vegetation.
The amount of hypothetical carbon emission that can cause the same change in global mean temperature as the temperature change due to biophysical forcings. It helps to quantify the temperature effects of biophysical forcings in terms of carbon. Negative carbon emission equivalent represents a carbon sink, suggesting a global cooling effect from the biophysical forcings.
The difference between C sequestration potential and C-emission equivalent as a C metric to assess the combined effect of biological carbon sequestration and biophysical forcings on temperature for forestry projects. It can serve as an index to quantify the climate regulation value of ecosystems.
An integrated quantification of climate regulation services in terms of C equivalents. The integration typically encompasses diverse factors, including biophysical forcings, fluxes of greenhouse gases, the carbon footprints of operations and energy costs, and carbon leakage from disturbances. Conversions of individual factors to carbon emissions often occur through the concept of global warming potential for non-CO2 greenhouse gases, as used in life-cycle analysis and other comparative frameworks.

May 2014

LAND USE, BIOPHYSICS, AND FORESTRY

333

generating the MODIS ET time series products (Mu et al. 2011). The CERES data included monthly averaged products of top of atmosphere (TOA)/surface longwave and shortwave fluxes and monthly gridded cloud products, with a spatial resolution of 18318 (Wielicki et al. 1998). The advanced microwave scanning radiometer-EOS (AMSRE) data included the five-day Level 3 global snow water equivalent (SWE) at 25-km resolution (Kelly et al. 2003).
For each product, we analyzed all available data for the most recent version as of December 2011. All products except MODIS ET were retrieved directly from satellite radiometric signals using dedicated algorithms; the MODIS ET product was derived using an empirical Penman-Monteith model by Mu et al. (2007) from meteorological and remote-sensing observed inputs such as air temperature and leaf area index. The theoretical basis, retrieval algorithms, and data validation for each product are available from the citations in the preceding paragraph. In particular, MODIS albedo and LST products have proven useful for characterizing biophysical variables of contrasting land surfaces at pixel scales (e.g., Montenegro et al. 2009). We also used two ancillary datasets: the 90-m digital evaluation model (DEM) from the Shuttle Radar Topography Mission (SRTM; data available online)8 and the ratio of diffuse/ direct downward surface shortwave fluxes calculated from a one-year model simulation using the coupled Weather Research Forecast-Community land model (WRF-CLM; Lu and Kueppers 2012).
Comparisons of surface biophysics between contrasting vegetation
MODIS products were used primarily to evaluate differences in surface albedo, LST, and ET between contrasting vegetation types (Table 1). Our evaluations emphasized paired adjacent sites (i.e., pixels) for two contrasting vegetation types. These adjacent sites were most likely to be found in ecotones and disturbed lands where the potential of future land-cover changes due to either natural processes or human activities is high. By using adjacent sites, we attempted to isolate vegetation controls on surface biophysics to the greatest extent possible and to minimize the influences of confounding factors such as topography, solar angle, and rainfall. Such comparisons are particularly relevant for land-use/ land-cover change because forestry conversions occur mostly at small scales. Differences between adjacent pixels/sites should also mimic future changes in surface biophysical characteristics with vegetation replacement.
Because we were interested mainly in comparing typical differences in biophysical variables for vegetation types, MODIS products such as albedo, ET, and LST were averaged across years from 2001 to 2011 for each of the 46 eight-day observation periods to smooth out interannual variability. In this averaging, the 11 MODIS quality-control flags associated with years 2001–2011 for
8 http:srtm.usgs.gov/index.php

each pixel and eight-day observation period were checked to select the years that had the best data quality; if the number of years with the best quality was less than five out of 11, we gradually incorporated the years with the next best quality; however, these years were assigned a weighting factor only half that of the higher quality years. This weighted-averaging procedure helped to reduce random errors and any quality-control issues in the MODIS products.
To determine the spatial distributions of vegetation, we derived a land-cover map at 500-m resolution by synthesizing the nine yearly 500-m MODIS land-cover layers for 2001–2008. A pixel was assigned a particular vegetation class only if the pixel was classified as this class with at least a probability of 0.50 for more than five out of nine years; otherwise, the pixel was discarded from our analysis. This filtering helped to suppress the confounding effects of classification errors and potential land-cover changes that occurred between 2001 and 2008. Additionally, the resultant 500-m base map was aggregated to 1-km resolution, with a 1-km pixel labeled as a vegetation class only if its four 500-m component pixels all belonged to the same class; otherwise, the pixel was discarded from our analysis. The 500-m and 1-km land-cover maps each contain a total of 17 land-cover types, but we considered only four vegetation types: CRO, GRA, ENF, and DBF. The two synthesized maps helped to derive vegetation-specific albedo and LST at 500-m and 1-km resolutions, respectively. However, vegetation-specific ET was derived based on the third map, the 2001 1-km MODIS Collection-4 land cover, because it is the reference map for generating MODIS ET (Mu et al. 2011).
The adjacent sites chosen to compare biophysical variables between contrasting vegetation types were determined based on the three land-cover maps using a customized local searching-window procedure. To suppress topographic influences, we considered only the sites with slopes of ,158. Using comparisons of albedo between DBF and CRO to illustrate this procedure, for each DBF pixel, all the CRO pixels within a 15-km radius of it were identified and a DEM filter was applied to select only those CRO pixels that had elevation differences ,10 m from the reference DBF pixel. The average albedo over all the final CRO pixels was then computed and compared against that for the reference DBF pixel. This local-scale comparison could also be performed using CRO pixels as the reference; our results showed that both potential choices of reference class gave essentially identical results. Additionally, this circular window of 30 km in diameter was moved across the study area to identify all the possible pairs of adjacent sites of contrasting vegetation.
Of the MODIS surface biophysical variables studied here, albedo depends not only on vegetation and soil properties, but also on solar angle and atmospheric conditions. To isolate the dependence of albedo on surface characteristics, we considered MODIS broad-

334

KAIGUANG ZHAO AND ROBERT B. JACKSON

Ecological Monographs

Vol. 84, No. 2

band white-sky albedo in our comparisons of paired sites. White-sky albedo is also called diffuse albedo, representing bi-hemispherical reflectance under isotropic skylight illumination; therefore, it is independent of sky conditions (Schaaf et al. 2002). ET is often constrained by water availability. Currently, the MODIS ET algorithm did not explicitly differentiate between irrigated and nonirrigated lands (Mu et al. 2011). However, the algorithm has synthesized information from some input variables responsive to soil water content; therefore, it indirectly captured the effect of irrigation such that irrigated lands in the MODIS ET often have high ET. To examine the spatial patterns of differences in albedo or ET, we applied the k-means algorithm to group the paired sites associated with each pair of vegetation types into three spatial clusters based on the similarity of seasonal variations in albedo or ET. More importantly, differences in biophysical variables between contrasting vegetation at the paired sites were used to evaluate biophysical forcings for potential vegetation shifts. These included shortwave and longwave radiative forcings and changes in surface sensible and latent heat fluxes, as detailed in the next four subsections.
Shortwave radiative forcing (SF) induced by albedo change
Altered surface albedo from land-use change induces shortwave RF, often evaluated at three levels: surface, atmosphere, and top of atmosphere (TOA). Specifically, RF at the surface (SFsfc) affects the surface energy balance and partitioning. RF at the TOA (SFtoa) is the quantity related to the change in global mean temperature through climate sensitivity parameters. Atmospheric RF (SFatm) is the difference between TOA and surface RFs (i.e., SFatm ¼ SFtoa À SFsfc); it represents the radiative imbalance of the atmospheric column and provides information on expected changes in precipitation and vertical mixing. Considerations of the vertical structure of RF have been recently requested for future climate assessments by the U.S. National Research Council (2005), although TOA RF still remains the most commonly used metric for quantifying and ranking the climatic impacts of different forcing agents. Calculations of TOA RF (SFtoa) require translating surface albedo (asfc), as measured by MODIS, to planetary albedo at the TOA (atoa), generally with asfc contributing to only a small fraction of atoa. This translation requires vertical profiles of atmospheric optical properties as determined by atmospheric compositions such as aerosol and cloud cover.
The single-layer radiative transfer model of Liou (2002) offers a simple yet effective scheme relating surface asfc to TOA albedo, atoa. This model uses two column-integrated optical parameters, namely, singlepass atmospheric reflectance R and transmittance T. We extended this model to further discriminate clear and cloudy skies within a grid as follows:

F"toa atoaðasfcÞ ¼

S

¼ cR þ ð1 À cÞR þ ca

Tc2ld

cld

clr

sfc 1 À asfcRcld

þ ð1 À cÞa

Tc2lr

ð1Þ

sfc 1 À asfcRclr

where S and F"toa are the incident and reflected solar fluxes at the TOA, respectively; c is the fraction of cloud

cover; Rcld and Tcld are the single-pass atmospheric reflectance and transmittance for the cloudy sky,

respectively; and Rclr and Tclr are for the clear sky. The sum of the first two terms cRcld þ (1 À c)Rclr is treated as the atmospheric contribution to TOA albedo

atoa, whereas the sum of the last two terms is the surface contribution to atoa. Accordingly, the relative fraction of surface albedo contributed to TOA albedo is calculated

as



T2

T2 .

casfc

cld þ ð1 À cÞasfc

clr

asfc :

1 À asfcRcld

1 À asfcRclr

Further, downward and upward shortwave fluxes

(sunlight) at the surface are given by





F# ðasfcÞ ¼ S 3 c Tcld þ ð1 À cÞ Tclr

sfc

1 À asfcRcld

1 À asfcRclr

F"sfcðasfcÞ ¼ asfcF#sfcðasfcÞ:

ð2Þ

The dependences of TOA and surface fluxes on albedo

asfc have been made explicit on the left-hand side of Eqs.

1 and 2.

Following a method similar to Donohoe and Battisti

(2011), we estimated the atmospheric reflectance R and

transmittance T of Eqs. 1 and 2 for both the clear and

cloudy sky portions monthly for each 18 3 18 grid, using

the CERES daytime cloud cover data and the CERES

cloudy-sky and clear-sky TOA/surface shortwave fluxes.

Our estimated atmospheric reflectance R and transmit-

tance T characterize the actual atmospheric optical

properties and allow us to compute surface, TOA, and

atmospheric shortwave RFs as follows:

h

i

SFtoa ¼ ÀS 3 atoaðasfc;2Þ À atoaðasfc;1Þ

SFsfc ¼ F#sfcðasfc;2Þ 3ð1 À asfc;2Þ À F#sfcðasfc;1Þ 3ð1 À asfc;1Þ

SFatm ¼ SFtoa À SFsfc:

ð3Þ

Here, the RFs are driven by a change in surface albedo from asfc,1 to asfc,2 while assuming that the atmospheric optical properties, including c, R, and T, remain unaffected. A positive shortwave RF in Eq. 3 indicates that the system absorbs extra solar radiation after land conversion. Eqs. 1–3 are applicable for computing instantaneous or short-time RFs that can then be integrated to estimate long-term RFs such as annual RFs.

May 2014

LAND USE, BIOPHYSICS, AND FORESTRY

335

FIG. 2. Schematic of the major biophysical forcings we examined: Land-use change by forestry alters the surface biophysics to induce both radiative and non-radiative forcings (left) that modify the cooling effect of forest carbon uptake (right). Radiative forcings, either shortwave (SF) or longwave (LF), perturb the radiation balance at the surface (sfc) and the top of the atmosphere (toa), or within the atmosphere column (atm). Non-radiative biophysical forcings exert strong controls on the redistribution of surface energy. In particular, enhanced evapotranspiration (ET) from forests lowers land surface temperature (LST) while a reduced input of sensible heat to the air tends to cool the near-surface air locally. The relative magnitudes of the competing effects of reduced albedo and carbon storage associated with reforestation and afforestation is often gauged by a metric called net carbon drawdown. In the schematic, gray-filled boxes denote components whose influences are not locally confined to the converted land. The equations we used are also labeled; klcc and kCO2 denote climate sensitivities for land use and CO2 changes, respectively. As an observation-based study, our analyses do not capture all the feedbacks and interactions between land and the atmosphere.

Using Eqs. 1–3, we calculated monthly RFs for four scenarios of non-forest to forest conversions (GRA or CRO to ENF or DBF) at a 18 3 18 resolution, compatible with the spatial and temporal resolutions of R and T derived from the CERES data. In the calculation, we considered actual surface albedo for asfc, which was estimated as the average of MODIS blacksky (direct) and white-sky (diffuse) albedos weighted by direct and diffuse downward fluxes from the regional climate simulation of WRF-CLM. Also importantly, albedos of adjacent sites as obtained from local comparisons of paired vegetation types were used in Eq. 3 for asfc,1 and asfc,2 to mimic realistic local vegetation shifts. The albedo values of all the paired sites within each 18 3 18 grid were averaged, and the resulting mean albedos were then applied to Eq. 3 for estimating the grid-level RF. Only those grids containing at least three pairs of adjacent sites were considered.
Carbon emission equivalent to albedo change
Albedo-induced shortwave RF at the TOA (SFtoa; in W/m2) is often converted to carbon emission equivalent

(dCalb; in kg/m2), a C density that can be compared against the C sequestration potential dCseq of land management to contrast biogeochemical and biophysical effects (Fig. 2). To date, the standard conversion method has generally overlooked the fact that the RFs from altered albedo and CO2 manifest different vertical structures, so that the same amount of RFs from these two forcing agents leads to different changes in global mean temperature (Betts 2000). Typically, these differing responses are characterized by climate sensitivity (k), defined as the change in global mean temperature per unit RF for a forcing agent and taken here as kalb ¼ 0.52 K/(W/m2) for albedo and kco2 ¼ 1.0 K/(W/m2) for CO2 (Davin et al. 2007). Accounting for this disparity may alter the conclusions of some earlier studies that assumed the same climate sensitivity for the two types of RFs in evaluating temperature benefits of reforestation.
We revised the standard method of converting RF SFtoa to carbon-emission equivalent dCalb by differentiating the two climate sensitivity parameters kalb and kco2. For a RF of SFtoa resulting from albedo change

336

KAIGUANG ZHAO AND ROBERT B. JACKSON

Ecological Monographs

Vol. 84, No. 2

over a local area slcc (m2), the total radiative perturbation is SFtoaÁslcc, which becomes SFtoaÁ(slcc/SE) if spread evenly across the globe with SE ¼ 5.1 3 1014 m2 being the total surface area of the Earth. Further, this global
shortwave RF is multiplied by the ratio of climate
sensitivity kalb/kco2 to obtain an effective CO2-induced longwave RF SFtoaÁ(slcc/SE)Ákalb/kco2. The ratio kalb/kco2 is often termed the climate efficacy (Hansen et al. 2005)
and is applied here to ensure that the two types of RFs
yield the same amount of global temperature response
according to their respective climate sensitivities. Then, the effective global longwave RF, SFtoaÁ(slcc/SE)Ákalb/ kco2, is converted to a change in atmospheric CO2 concentration dCco2 (parts per million per volume [ppmv]) via the efficiency parameter of 5.35 W/m2 as
follows:
0 kalb slcc 1 1 dCco2 ¼ [email protected] 3 kco2 3 SE 3 5:35 À 1CA 3 Cco2 ð4Þ

where Cco2 ¼ 391 ppmv is the reference CO2 concentra-
tion. Finally, the CO2 change dCco2 is translated to the land C emission dCalb (kg/ m2) over the area slcc as
follows:

dCalb ¼ j 3 dCco2 1 ’ Cco2 3 j 3 SFtoa

0:50 slcc

0:50 3 5:35 3 SE

¼ 0:611 3 kalb 3 SFtoa

ð5Þ

kco2

where the constant of 0.50 in the denominator is the airborne fraction, representing the portion of C emission that remains in the air after being absorbed by the ocean and other terrestrial sinks (Betts 2000, Montenegro et al. 2009), and j (2.13 3 1012 kg/ppmv) is the coefficient converting C from ppmv to kg. The term dCalb refers to the emission of carbon and can be converted to a CO2 equivalent by multiplying by 3.67.
The C emission equivalent dCalb converted from SFtoa can be used to adjust the actual C sequestration potential of forests dCseq, which results in a net carbon drawdown metric (i.e., dCseqÀalb ¼ dCseq À dCalb) for quantifying the temperature benefit of forestry projects. The value of dCalb is typically positive for conversions to forests because of the reduced albedo and increased shortwave absorption in consequence; it represents the minimum C uptake that the trees need to sequester compared to the replaced vegetation for offsetting the warming of reduced albedo. Therefore, positive net carbon drawdown (i.e., dCseqÀalb . 0) indicates a global cooling in terms of the integrated effects of albedo reduction and CO2 uptake. Calculating net carbon drawdown dCseqÀalb requires the actual C sequestration potential dCseq, a quantity often estimated as the difference in steady-state C stocks before and after land conversion. However, to our knowledge no reliable data sets are available for spatially explicit mapping of C stocks of different vegetated lands at a scale commen-

surate with the satellite data we used, especially for belowground carbon. Moreover, the definitions and calculations of carbon sequestration, dCseq, for forestry projects varied considerably among prior studies, particularly concerning how the studies treated landuse history and forest management practices. Therefore, we did not estimate exact values of dCseq or dCseqÀalb, but just inferred the potential signs of net carbon drawdown dCseqÀalb by referring to previous estimates of approximate C sequestration potential, dCseq, of forestry projects (Betts 2000, Claussen et al. 2001, Gibbard et al. 2005, Montenegro et al. 2009).
Longwave radiative forcing induced by changes in surface temperature/emissivity and ET
Altered surface biophysical properties modify the longwave radiative regime through at least two mechanisms, one pertinent to LST and emissivity and another to ET (Fig. 2). Specifically, surface longwave RF from the altered LST and emissivity is defined as the change in net downward longwave radiation at the surface and is calculated by
LFsLfScT ¼ ðe1r1T14 À e2r2T24Þ þ L#ðe2 À e1Þ ð6Þ
where T1 and T2 denote LSTs before and after land conversion, with the corresponding emissivity being e1 and e2; L# is the downward sky longwave flux and is assumed to be unchanged; and r (5.67 3 10À8 WÁm2ÁKÀ4) is the Stefan-Boltzmann constant. Of the two surface terms in Eq. 6, the first, attributable mainly to the temperature change, dominates, whereas the second, attributable to the emissivity change, is small and often negligible. A positive value of LFLsfScT indicates the suppression of thermal emission after the land conversion or, expressed differently, less longwave radiation dissipated from the converted surface attributable to its lowered LST. This suppression also decreases both the longwave radiation absorbed by the atmosphere and that escaping at the TOA. Globally, only an average of 22 W/m2 surface emission out of 390 W/m2 (;5.6%) escapes into space (Costa and Shine 2012), a value lower than the previous estimate of 40 W/ m2 (;10%) by Trenberth et al. (2009). This global-scale partitioning provides ratios to roughly apportion surface longwave forcing LFsfc into longwave RFs at the TOA and for the atmosphere as follows:
LFLtoSaT ¼ 22=390 3 LFLsfScT

LFLatSmT ¼ À368=390 3 LFLsfScT

ð7Þ

where the ‘‘minus’’ sign in the apportioning for LFatm ensures that a positive LF value means that the system gains more longwave radiation or loses less radiation compared to the original vegetation. In our analyses of the four classes of vegetation replacement, LST and downward longwave flux as used in Eq. 6 are obtained from the MODIS daytime and nighttime LSTs and the

May 2014

LAND USE, BIOPHYSICS, AND FORESTRY

337

CERES longwave flux product, respectively. Surface

emissivity is estimated from albedo using an empirical

relationship developed by Juang et al. (2007), e ¼ 0.99 –

0.16asfc.

The second type of longwave RF is caused by the

change in atmospheric water vapor concentration due to

altered ET. Different from the albedo-induced RF of

Eq. 3 or the LST-induced longwave RF of Eq. 6, this

type of RF is not necessarily confined to locations where

ET is altered, attributable to the dynamic nature of

atmospheric water vapor transport, which makes its

computation difficult. To provide rough estimates for

ET-induced longwave RF only, we considered an

extreme case assuming that the water vapor is well

mixed at the global scale. This assumption allows us to

calculate this RF, LFEtoTa, for a given time t in a way
similar to that of CO2, but with a different forcing parameter of 20.7 W/m2, as follows:





LFETðtÞ ¼ 20:7log 1 þ slcc 3 DmH2OðtÞ 3 SE

toa

MH2O

slcc

¼ 0:83 3 DmH2OðtÞ

ð8Þ

where MH2O is the total mass of water vapor in the troposphere, taken simply as 1.27 3 1016 kg; slcc and SE

again are the respective areas of the converted land and

the Earth; the value of 20.7 for the forcing parameter is

derived from the simulation results of Collins et al. (2006); and DmH2O (kg/m2) is the cumulative change in

the atmospheric water vapor at time t contributed by the

altered ET input per unit area. This water vapor change

is estimated by assuming a mean residence time of 10

days for water vapor:

Zt





ðt À sÞ

DmH2O ¼ 0 DETðsÞ 3 exp À 10 ds ð9Þ

where the change in ET at a given time s (day) for a land conversion, DET(s) (kg H2OÁmÀ2ÁdÀ1; hereafter expressed as kgÁmÀ2ÁdÀ1), is obtained from the MODIS
ET comparison based on adjacent sites. In Eq. 8, SE/slcc is applied to transform the global mean ET-induced longwave forcing (i.e., 20.7 Á log(1 þ slccÁDmH2OðtÞ/MH2O)) to its effective local value LFEtoTa to be comparable to the local shortwave RF SFtoa. Our estimate of LFEtoTa is approximate; we have not attempted to translate this
ET-induced longwave forcing into an equivalent C
emission.

Non-radiative forcings and re-partitioning of surface energy associated with land conversion
Non-radiative forcings associated with changes in biophysical properties, such as surface roughness, canopy conductance, canopy structure, and rooting depth, also affect temperature. We examined the local impact of these forcings at the surface through the redistribution of sensible (H ) and latent (kÁET) heat (Fig. 2). In our monthly or annual analyses, the downward energy flux into the ground is typically small;

thus, the surface energy balance equation becomes Rn ¼ H þ kÁET or DRn ¼ DH þ kÁDET. Here, DRn represents the change in surface net radiation after land conver-
sion, and it is dominated by LST- and albedo-induced RFs, DRn ¼ SFsfc þ LFLsfScT, because the contributions of ET- and CO2-induced longwave RFs to the local energy balance are close to zero. The change in latent heat flux kÁDET was obtained from the adjacent comparisons of
MODIS ET using the heat of vaporization k as the
conversion factor. Then, the change in sensible heat flux
DH was estimated as

DH ¼ SFsfc þ LFsLfScT À k Á DET:

ð10Þ

Because the warming of near-surface air is fueled directly by sensible heat, we expect that a land conversion with increased sensible heat (DH . 0), regardless of the signs of DRn and DET, would tend to warm the planetary boundary layer locally and that conversely, a conversion with a negative DH would tend to cool the near-surface air locally. For example, an increase of 1 W/m2 in the sensible flux for a heating cycle of 12 h raises the temperature of a 250-m mixed layer by as much as 0.14 K, which is estimated approximately according to the simple formula of West et al. (2011). In contrast, enhanced latent heat (i.e., kÁDET . 0) does not immediately warm the near-surface air, even though this extra energy will be turned into sensible heat somewhere in the upper air, when condensing, and thus, will modify the energy balance of the atmosphere overall. This extra latent heat impacts the local or regional surface energy balances through indirect pathways, such as the greenhouse effect of the associated water vapor and the attenuation of sunlight if the water vapor condenses into cloud droplets. Such interactions are difficult to track directly from MODIS data. Rather, we referred mainly to ET-induced longwave RFs as a metric for assessing the potential impacts of altered ET on regional and global temperature.
The direct heating or cooling of the local atmospheric column above a disturbed land area is determined by both RF RFatm ¼ SFatm þ LFatm and non-radiative forcing DH. Unlike the change in sensible flux DH, the atmospheric radiative forcing RFatm affects temperature throughout the air column, with the maximum influence expected to occur in the middle layer, although the exact altitude depends on the atmospheric opacity at the respective spectral bands. Therefore, the direct effects of the atmospheric RFs SFatm and LFatm on the nearsurface air temperature are negligible compared to the non-radiative forcing DH.

RESULTS
Albedo
Latitudinal and seasonal variations in albedo resembled changes in snow-water equivalent (Appendix: Figs. A2 and A3), implying the critical role of snow in determining surface albedo (Figs. 3 and 4). Lands

338

KAIGUANG ZHAO AND ROBERT B. JACKSON

Ecological Monographs

Vol. 84, No. 2

FIG. 3. Comparisons of zonally averaged MODIS whitesky albedo among four land-cover types, including grassland (GRA), cropland (CRO), evergreen needleleaf forest (ENF), and deciduous broadleaf forest (DBF), along the latitude range of 208–608 N for a winter (top) and a summer (bottom) around day of the year 17 and 233, respectively. The zonal averaging was performed using a 0.18-latitude bin for zones with more than five MODIS pixels of a vegetation class. For comparison, the associated advanced microwave scanning radiometer-EOS (AMSR-E) snow water equivalent (SWE) is also depicted. Note the different y-axis scales in January vs. August.
covered with herbaceous plants or short vegetation normally had brighter surfaces than lands with woody vegetation throughout the year, especially when snow is present (Fig. 5). The zonally averaged MODIS whitesky albedo over 458–608 N around day of year 17 in January, for example, was 0.57 6 0.05, 0.50 6 0.07, 0.26 6 0.04, and 0.20 6 0.02 (mean 6 SE) for CRO, GRA, DBF, and ENF, respectively. Unlike multilayered forest canopies, croplands and grasslands, which are covered with little or no low-lying live or dead biomass in winter, are more likely to be buried under snow. Foliage losses in deciduous forests exposed more snow-covered ground than in evergreen forests, enhancing the observed wintertime albedo of DBF somewhat compared to ENF (Figs. 3 and 4).
Paired local comparisons of adjacent vegetation types show consistent differences in albedo (Fig. 5). Albedos of CRO and GRA were, on average, greater than those of ENF and DBF. For instance, albedos of GRA and

ENF differed by 0.21 in January and 0.054 in July when averaged over paired sites (P , 0.001, n ¼ 317 911 paired sites), representing an increase of 97% and 51%, respectively. Croplands generally had albedo values similar to those of adjacent grasslands, although croplands on average were slightly brighter (i.e., an annual mean albedo of 0.216 vs. 0.211 averaged over all the CRO–GRA sites, P , 0.0001, n ¼ 2 037 020). At most of the paired sites, ENF had lower albedo than DBF throughout the year (i.e., annual mean albedo of 0.138 vs. 0.166, P , 0.0001, n ¼ 103 766); thus, of the two forest types, DBF tended to have albedo closer to that of CRO or GRA. Another observed pattern was that the four types of vegetated surfaces in temperate regions, especially DBF, all showed a gradual increase in albedo at the beginning of snow-free seasons, and then a gradual decline before snowfall in autumn (Fig. 4). This pattern is driven primarily by seasonal foliage dynamics.
The magnitude of albedo differences between adjacent vegetation types varied with location, as indicated in the results from the k-means clustering (Fig. 6). The resultant clusters correspond to distinct geographic regions and were determined mainly by wintertime albedo, reflecting the differing regimes of snow and vegetation interactions across regions. For example, the three clusters of the DBF–CRO sites occupied distinct latitudinal bands in the Eastern USA (Fig. 6); thus, latitude can serve as a proxy to explain the observed pattern in albedo difference between DBF and CRO. As another example, the difference in annual albedo between CRO and GRA was 0.005 (P , 0.001) when averaged over all the paired CRO–GRA sites, compared to those of À0.0007 (P ¼ 0.012), 0.02 (P , 0.001), and À0.036 (P , 0.001) when averaged separately over the three clusters (Fig. 6): This spatial pattern suggests some differences in wintertime standing biomass of grasses and crops across regions, which affect albedo dynamics of snow-covered surfaces and are caused in part by differences in crop types and management practice. The pattern may also be influenced by the spatial variation in amount and length of snow cover.
Land surface temperature (LST)
The seasonal and latitudinal variations of land surface temperature (LST) were determined by both the incoming TOA solar radiation and land surface characteristics. The percentages of spatiotemporal variations in LST explained by the TOA solar radiation were 79.4%, 82.8%, 67.2%, and 82.6% for ENF, DBF, GRA, and CRO, respectively (Appendix: Fig. A4). The unexplained variations partially underscore the effects of surface characteristics on LST. In terms of zonally averaged summertime LST, GRA often appeared to be the warmest, followed by CRO, ENF, and DBF. For example, at 358 N, GRA surfaces were ;5.0 K warmer than ENF in July. Our paired local comparisons further reveal the apparent controls of vegetation on LST (see Fig. 7). In terms of daily LST, forested surfaces were