View metadata, citation and similar papers at core.ac.uk brought to you by CORE

provided by UNH Scholars' Repository

University of New Hampshire University of New Hampshire Scholars' Repository Institute for the Study of Earth, Oceans, and Space Earth Systems Research Center (EOS)

1-1-2016 Tradeoffs between three ecosystem services across the state of New Hampshire, USA: Timber, Carbon, and Albedo David A. Lutz Dartmouth College

Elizabeth A. Burakowski University of New Hampshire, Durham, [email protected]

Mackenzie B. Murphy Dartmouth College

Mark E. Borsuk Dartmouth College

Rebecca M. Niemiec Dartmouth College

See next page for additional authors

Follow this and additional works at: https://scholars.unh.edu/ersc

Recommended Citation Lutz, DA, EA Burakowski, MB Murphy, ME Borsuk, RM Niemiec, and RB Howarth. 2016. Tradeoffs between three forest ecosystem services across the state of New Hampshire, USA: Timber, Carbon, and Albedo. Ecological Applications 26(1): 146-161. https://dx.doi.org/10.1890/14-2207

This Article is brought to you for free and open access by the Institute for the Study of Earth, Oceans, and Space (EOS) at University of New Hampshire Scholars' Repository. It has been accepted for inclusion in Earth Systems Research Center by an authorized administrator of University of New Hampshire Scholars' Repository. For more information, please contact [email protected]. Authors David A. Lutz, Elizabeth A. Burakowski, Mackenzie B. Murphy, Mark E. Borsuk, Rebecca M. Niemiec, and Richard B. Howarth

This article is available at University of New Hampshire Scholars' Repository: https://scholars.unh.edu/ersc/115 Ecological Applications, 26(1), 2016, pp. 146–161 © 2016 by the Ecological Society of America

Trade- offs between three forest ecosystem services across the state of New Hampshire, USA: timber, carbon, and albedo

D AVID A. LUTZ , 1,6 E LIZABETH A. BURAKOWSKI , 2,3 M ACKENZIE B . M URPHY ,4 M ARK E. BORSUK ,4 R EBECCA M. NIEMIEC , 1,5 AND R ICHARD B . H OWARTH 1 1Environmental Studies Program, Dartmouth College, Hanover , New Hampshire 03755 USA 2National Center for Atmospheric Research, Climate and Global Dynamics Division, Boulder , Colorado 80305 USA 3 Earth Systems Research Center, Institute for the Study of Earth Oceans, and Space, University of New Hampshire, Durham , New Hampshire 03824 USA 4Thayer School of Engineering, Dartmouth College, Hanover , New Hampshire 03755 USA 5Emmett Interdisciplinary Program in Environment and Resources, School of Earth Sciences, Stanford University, St anford, California 94305 USA

Abstract . are more frequently being managed to store and sequester carbon for the purposes of climate change mitigation. Generally, this practice involves long- term conservation of intact mature forests and/or reductions in the frequency and intensity of timber harvests. However, incorporating the infl uence of forest surface albedo often sug- gests that long rotation lengths may not always be optimal in mitigating climate change in forests characterized by frequent snowfall. To address this, we investigated trade-offs between three ecosystem services: carbon storage, albedo- related radiative forcing, and timber provisioning. We calculated optimal rotation length at 498 diverse and Analysis forest sites in the state of New Hampshire, USA. We found that the mean optimal rotation lengths across all sites was 94 yr (standard deviation of sample means = 44 yr), with a large cluster of short optimal rotation lengths that were calculated at high elevations in the White Mountain National Forest. Using a regression approach, we found that timber growth, annual storage of carbon, and the difference between annual albedo in mature forest vs. a post- harvest landscape were the most important variables that infl uenced optimal rotation. Additionally, we found that the choice of a baseline albedo value for each site signifi cantly altered the optimal rotation lengths across all sites, lowering the mean rotation to 59 yr with a high albedo baseline, and increasing the mean rotation to 112 yr given a low albedo baseline. Given these results, we suggest that utiliz- ing temperate forests in New Hampshire for climate mitigation purposes through carbon storage and the cessation of harvest is appropriate at a site-dependent level that varies signifi cantly across the state. Key words: albedo ; climate mitigation ; ecological economics ; ecosystem services ; forest carbon storage ; ; ; temperate forests ; White Mountain National Forest, New Hampshire, USA

I NTRODUCTION regulatory frameworks, for instance the United Nation ’ s Clean Development Mechanism (Galik et al. 2014 ) Roughly 27% of anthropogenic emissions of carbon and the California Environmental Protection Agency’ s dioxide are sequestered in terrestrial ecosystems (Le Air Resources Board, as well as through voluntary Quéré et al. 2013 ), and the majority of this storage mechanisms (Hurteau et al. 2013 ). A large focus of occurs in forested landscapes where it is stored in forest climate mitigation strategies has focused stem and biomass. Markets for forest-based on tropical (Gibbs et al. 2007 ), yet as of carbon offsets have been designed to take advantage late, domestic funding of such strategies in Europe of this potential carbon sink through both formal and North America have garnered interest in temper- ate forests as well (Tittmann and Yeh 2013 ). In response Manuscript received 21 November 2014 ; revised 10 to these incentives, temperate forest ecosystems are April 2015 ; accepted 21 April 2015. Corresponding Editor: A. D. McGuire. increasingly being managed to provide such climate- 6 E-mail: [email protected] regulating ecosystem services (Canadell and Raupach

146 January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 147

2008 , Fahey et al. 2010 , Law and Harmon 2011 ). As The most recent developments in calculating the a result, understanding how forest management infl u- overall balance between carbon storage and albedo in ences the total storage of forest carbon has been a forested landscapes have been the result of careful particular point of interest in the research and man- empirical models of forest ecosystems as informed by agement communities (Keith et al. 2014 ). forest growth data coupled with albedo measurements Previous modeling studies of boreal and temperate from satellite sensors (Bright et al. 2011 , 2012 , 2014 , forests in the scientifi c literature have generally indi- Cherubini et al. 2013 ). Generally these studies have cated that incorporating the value of carbon sequestra- occurred in boreal forests, where fi ndings suggest that tion leads to long optimal rotation lengths (van Kooten species cover, particularly whether the forest is broad- et al. 1995 , Kaipainen et al. 2004 , Price and Willis leaved deciduous or evergreen coniferous, as well as 2011 ). As prices for carbon increase, rotation lengths stand age, temperature, and latitude (Bright et al. 2013 , also increase, with complete conservation eventually Lukeš et al. 2013 ) infl uence surface albedo. While these being realized at high prices (Chladná 2007 , Gutrich studies have relied upon the suite of MODIS albedo and Howarth 2007 , Liski et al. 2011 ). While there are products, MOD10A (Hall and Riggs 2007 ), and methodologies to incorporate partial and limited harvest MCD43A (Schaaf et al. 2002 ), empirical models of so as to increase net primary production and rates albedo dynamics have been generated (Bright et al. of carbon storage temporarily (Hardiman et al. 2013 ), 2013 ) and used (Lukeš et al. 2013 , 2014 , Otto et al. for the most part these efforts have a small infl uence 2014 ) specifi cally for this purpose. Although a few on total carbon storage. Subsequently, most forest studies have used similar methods to examine temper- carbon-offset projects are designed around the con- ate forests (Thompson et al. 2009 , Lutz and Howarth servation of intact forests and the elimination of most 2014 ), these efforts looked at broad categories of forest harvest. type and did not examine enough sites to understand A wrinkle in the existing forest carbon conservation subtle variations across the landscape. It is important paradigm comes from a well- established line of research to note that, while trade- offs between these biogeo- indicating that forest ecosystems interact with the physical forcings have been estimated before at coarse atmosphere through biogeophysical mechanisms in (e.g., Zhao and Jackson 2014 ) and fi ne scales (Williams addition to gaseous exchange (Bonan et al. 1995 , Bonan et al. 2014 ), these studies have not included method- 2008 , Jackson et al. 2008 ). In northern latitudes, boreal ologies to relate forcings to each other using a similar and temperate forest ecosystems that are frequently unit of analysis or value. covered by snowfall have been found to infl uence cli- The calculation of net benefi ts from carbon storage mate substantially through surface- albedo-related radia- in forested ecosystems is generally based on an expected tive forcing; in some cases, this forcing outweighs that baseline of forest growth (Malmsheimer et al. 2011). of storage sequestration in forest biomass (Betts 2000 , From this baseline, calculations can be made to ensure Chapin et al. 2000 , Kirschbaum et al. 2011 ). This is that management actions have generated a benefi t (i.e., mainly thought to be due to the low albedo of conifer a net increase in carbon storage) when compared to needle-leaf tree species (Sturm 2005 ), as well as low a situation when no actions were taken. This is defi ned growth rates and reduced carbon storage in the forests by the term “additionality” and is a requisite for cli- themselves (Lutz and Howarth 2014 ). As a result, the mate change policy frameworks. Calculating national existing mechanism for mitigating climate change in baselines of forest growth is a contentious process forests, conservation and the elimination of harvest, with serious consequences for the total credited emis- may not necessarily generate a cooling infl uence on sions allocated to forest projects (Griscom et al. 2009 ). global climate when albedo is considered (Bright et al. Previous studies that have included trade- offs between 2014 ). Conversely, the management mechanism for albedo and carbon storage have used available satellite increasing albedo requires the removal of forest bio- records of albedo (Bright et al. 2012 , 2014 , Lutz and mass, which counteracts the attempt to store carbon Howarth 2014 ), yet it is uncertain whether or not in woody biomass. In order to properly and optimally that is a statistically reliable sample of measurements design effi cient and effective climate change mitigation upon which to categorize future behavior of albedo. policies, it is therefore important to understand how Thus, the choice of which annual albedo values are these two climate- regulating ecosystem services trade selected to compare to future albedo, what we deem off under different management regimes (Thompson an albedo baseline, is critically important for the proper et al. 2009 , Anderson et al. 2011 , Bright et al. 2011 , implementation of albedo into climate policy. 2012 ). However, as forests are complex ecosystems In this study, we seek to address this lack of under- that vary in species composition, growth rate, and standing among the trade-offs between multiple eco- carbon storage over space and time, a detailed under- system services in New Hampshire forest landscapes standing of the exact trade- offs between the benefi ts by focusing on three related questions: (1) Is the of carbon storage and increased albedo- related radiative method of reducing timber harvest in forests an opti- forcing remains elusive for many temperate and boreal mal strategy in general for climate mitigation across forests. diverse temperate forests? (2) If not, under what 148 DAVID A. LUTZ ET AL. Ecological Applications Vol. 26, No. 1 circumstances should harvest for albedo- induced radia- encompassing over a dozen dominant forest species tive cooling or lack of harvest for common throughout the New England region (Fig. 1 ). be emphasized? (3) Does the selection of an albedo The dominant forest type in New Hampshire is baseline have a signifi cant infl uence on the optimal categorized as northern hardwood, which usually rotation length when timber, albedo, and carbon are contains a combination of maples (Acer sacccharum considered? and A. rubrum ), beech (Fagus grandifolia ), birch (Betula We hypothesize that although most temperate alleghaniensis and B. papyrifera ), and occasionally ash forest climate mitigation projects are focused on a (Fraxinus americana ) and Eastern hemlock (Tsuga limited-harvest approach, this methodology may not canadensis). In the northern region of the state and be optimal when albedo is included in ecosystems at higher elevations, particularly in the White that receive frequent snowfall. We also expect that Mountain National Forest, the spruce–fi r forest, con- the ratio between forest growth rate and post-harvest taining red spruce (Picea rubens) and balsam fi r ( Abies albedo is the key metric through which to determine balsamea), dominates the landscape. In southern New the best harvest rotation length. Finally, we predict Hampshire, pine–oak forests, containing white pine that year- to- year variation in measured albedo may (Pinus alba), red pine (P. resinosa), red oak (Quercus dramatically impact optimal rotation length when rubrum ), and occasionally other hardwoods are pre- alternative albedo baselines are selected for use sent, particularly in areas once used for agriculture in the calculation of site net present value (NPV). or pasture. Hemlock forests, dominated by Eastern Collectively, our fi ndings have important implications hemlock, and aspen–birch forests, containing two for guiding the direction and design of climate miti- species of aspen (Populus grandidentata and P. tremu- gation projects in the state’ s forests and throughout loides), are less common forest types found mainly the New England region. in riparian valleys. In addition to the dominant forest species that occur in these broad major forest types, New Hampshire contains an additional 71 tree spe- M ETH ODS cies, many of which are uncommon but occasionally will become a signifi cant part of the dominant canopy, Study area and sites namely gray birch (B. populifolia ), black cherry (Prunus Our study examined a network of 498 forest inven- serotina), green ash (F. pennsylvanica), striped maple tory plots throughout the state of New Hampshire, (A. pensylvanicum ), and American elm (Ulmus USA, which represent a variety of forest types, americana ).

1 – 300 yr Albedo MODIS MOD10A/ rotations calculations MCD43A Albedo

Land cover type Daily radiative forcing MCD12Q1 FACT NPV

Radiative forcing 2003–2013 Albedo Carbon baseline Timber

Albedo NPV

DICE model Carbon NPV Total NPV $

Timber NPV

Carbon storage Timber revenues Grid search

FACT carbon FACT timber equations equations Maximum NPV, Optimal rotation period FVS-NE model 498 FIA sites

FIA stand Biological, biophysical site Regression Information variables tree

F IG. 1. A graphical representation of the methodologies used in this study. Each of the 498 FIA (Enhanced Forest Inventory and Analysis) sites in New Hampshire (USA) was simulated 299 times, with rotation lengths ranging 1–300 yr. Models include the FACT (Forest Albedo Carbon and Timber) and FVS-NE (Northeast Variant of the Forest Vegetation Simulator), inputs include net present value (NPV; US$) calculations; Δ refers to change in, Σ to sum of. January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 149

The forest plots used in this analysis were inventoried 1989 ). FVS-NE has been validated for northern hard- through the Enhanced Forest Inventory and Analysis wood forest ecosystems (Yaussy 2000 ), as well as for (FIA) Program of the U.S. Department of Agriculture simulating the growth of forests post-management (USDA) and the Forest Service (Bechtold and Patterson (Crookston and Dixon 2005 , Ray et al. 2009 ). An 2005 ). Through the FIA program, a systematically analysis of the FVS-NE in predicting the total tree scattered series of forest plots are inventoried annually height and 5-yr diameter and height increment for the throughout the United States. Sampling design for the 20 most abundant species in the northeastern United FIA program is constructed in a way that results in States found that the model ’ s estimations performed a nationwide sampling distribution of one plot for moderately well and did not show any major spatial every 2428 ha (24.28 km 2) in areas classifi ed as forests issues (Russell et al. 2013 ). (Bechtold and Patterson 2005 ). Additionally, the sam- For each site, we simulated forest growth for 200 pling design is constructed so as to reduce bias in yr at increments of 5 yr, assuming static present- day tree species distributions (Bechtold and Patterson 2005 , climate. At the fi rst year of each simulation, we pre- McRoberts et al. 2005 , Zhu et al. 2012 ). Forest plots scribed the removal of all greater than 2 inches were constructed through a four- point cluster design (1 inch = 2.54 cm) dbh, with no additional harvest containing four subplots of a radius of 7.32 m, wherein for the remaining time period. We selected this dbh trees 12.7 cm in diameter at breast height (dbh; 1.37 limit as it conforms to USDA guidelines for group m above ground) are measured and recorded; four selection harvest (Lamson and Leak 2000 ) and is com- 2.07 m fi xed- radius microplots in which saplings and monly used in other clear- cut simulations using FVS seedlings are measured and recorded; and four 18.0 (Nunery and Keeton 2010 , Schwenk et al. 2012 ). m fi xed-radius macroplots, which are used when large Modeling regeneration in FVS-NE requires careful trees are encountered (Bechtold and Patterson 2005 ). tuning and necessitates incorporating natural regenera- For our analysis, we initially relied on a base collec- tion parameters for simulations including harvest. For tion of 2083 forest plots collected from 2003–2012. this information, we relied upon previously published We removed plots that were defi ned as unstocked with fi eld data from several sources (Leak 2005 , Nunery timber, contained too few records to model accurately, and Keeton 2010 ) and used in other studies using or were not located in an area in which suffi cient FVS- NE simulations (Nunery and Keeton 2010 , Mika satellite data for the estimation of albedo were avail- and Keeton 2014 ) and followed the approach of Mika able, and thus refi ned our analysis to a total of 498 and Keeton (2014 ). We collected individual stand reports plots. for every 5 yr of the simulations, which included the simulated number of individuals, their sizes and ages, species, and position in the canopy. Additionally, stand Forest growth modeling structural attributes such as canopy closure and stand For each of our sites, we utilized the Northeast biomass were collected along with estimates of carbon Variant of the Forest Vegetation Simulator (FVS- NE) storage based on allometric equations specifi c to spe- model (Wykoff et al. 1982 , Dixon 2002 , Crookston cies groups (Jenkins et al. 2003 ). We did not incor- and Dixon 2005 ) to simulate growth following a clear- porate changes in climate or natural disturbance through cut harvest for a period of 200 yr. The FVS model windthrow in our forest simulations. is a distance-independent individual-tree growth model, which generally relies upon incremental diameter growth Timber and carbon modeling in large trees within each plot (Crookston and Dixon 2005 ). Diameter growth in FVS is infl uenced by several To calculate the total quantity of timber volume site characteristics including aspect, elevation, and from each site, we used the Forest Albedo Carbon general habitat type and further constrained by crown and Timber (FACT) model (Gutrich and Howarth position and other structural attributes which determine 2007 , Lutz and Howarth 2014 ) to estimate quantities the overall growth rate, mainly stem density (Crookston of pole- and sawtimber in each stand. The FACT and Dixon 2005 ). Small trees are simulated by a height- model relies upon the curvilinear relationship that exists growth approach, a methodology that calculates incre- between harvestable timber volume ( V) and stand age mental diameter growth based upon height growth as ( ) { ( )(s−𝛼 ) well as site characteristics such as a crown competition 𝛼 − −𝛼 2 s ≥ 𝛼 V (s) = 0 1 1 1 2 factor, stand density, crown ratio, and social position slt𝛼 0 2 (1) of the individual (Crookston and Dixon 2005 ). FVS α uses species specifi c allometric equations for biomass where 0 is the maximum timber volume for the site, α α estimates (Jenkins et al. 2003 , Nunery and Keeton 1 is the volume growth rate, 2 is the youngest stand 2010 ) and thus can be used to simulate above- and age when a stand can be harvested and contain valu- belowground carbon dynamics through time. able timber, and s is the stand age (Gutrich and Howarth Specifi cally, FVS-NE relies upon growth and yield 2007 ). For each stand, we calculated this relationship equations from the NE- TWIGS model (Hilt and Teck and its parameters based on the volumetric outputs 150 DAVID A. LUTZ ET AL. Ecological Applications Vol. 26, No. 1 of FVS-NE post-harvest through a process of optimi- its variable from site to site (Schwenk et al. zation described in the Appendix and based on the 2012 , Mika and Keeton 2014 ) and frequent lack of methods of Gutrich and Howarth (2007 ). We divided inclusion in life- cycle analysis and offset protocols (Law timber volume into sawtimber using the equation and Harmon 2011 ). Carbon stored in long- lived wood 𝛽 s 0 −𝛽 lt products was also modeled in a similar fashion to ⎧ 0 (s+𝛽 ) 2 0 1 Gutrich and Howarth ( 2007 ). ⎪ ( ) 𝛽 s [ ] f (s) = 𝛽 s∕ s+𝛽 −𝛽 0 −𝛽 ∈ ⎨ 0 1 2 s+𝛽 2 0,1 saw ( 1) ⎪ 𝛽 s ⎩ 0 −𝛽 > Albedo data collection 1 (s+𝛽 ) 2 1 1 (2) For each site, we collected snow-free surface albedo β β β in which 0 , 1 , and 2 are coeffi cients, with fsaw from the MODIS Bi- Directional Refl ectance representing the proportion of timber that is sawtimber, Distribution Function (BRDF) Adjusted Albedo prod- and poletimber representing the remaining fraction of uct (MCD43A3, v005; Schaaf et al. 2011 ) from 2002– timber volume (Gutrich and Howarth 2007 , Lutz and 2013. Specifi cally, we used the 500- m spatial resolution Howarth 2014 ). 8-d broadband combined Aqua and Terra product. To represent the appropriate costs associated with High-quality and snow- free retrievals of directional short rotation lengths, we implemented mowing costs α hemispherical refl ectance ( ksky) and bihemispherical that occurred when there was a timber harvest, yet α refl ectance ( wsky ) albedos were linearly interpolated when no pole- or sawtimber stock was present at the α to actual ( bsky ) albedo assuming an isotropic distribu- site (i.e., when the harvest time, t, was less than the tion of diffuse skylight (SKYL) at local solar noon α minimal stand age with timber volume, 2 ). Costs for through the equation mowing and maintaining fi elds in the New England 𝛼 = ×𝛼 +( − )×𝛼 region have been estimated to be between US$80 and bsky SKYL wsky 1 SKYL ksky (4) US$486 per ha (Oehler 2003 ), depending on site con- where SKYL has been set to 0.2. A sensitivity ditions. To account for this, we reasoned that as a analysis performed using SKYL values ranging from site became overgrown, the cost per clearing increases 0.1 to 0.6 did not yield any signifi cant differences and as such, we estimated that mowing costs, Mc , α in bsky computed from Eq. 3, thus suggesting that increased proportionately with stand age s , to a maxi- the simple linear interpolation captures the majority α mum value at 2 : α of the anisotropic dependence of bsky on aerosol ( ) optical depth and solar zenith angle for snow- free { s LC +0.835391 ×HC slt𝛼 surface albedo. M = 𝛼 −1 2 c 2 The number of high- quality snow- covered pixels for 0 s ≥ 𝛼 (3) 2 the MCD43A3 product was very limited (and in some where LC and HC are the low and high, respectively, cases zero) for many of the sites in our analysis. As estimates of mowing cost per hectare adjusted for such, we elected to use the MOD10A1 500- m daily α infl ation and 2 represents the youngest age at which snow- covered shortwave broadband albedo product the stand will contain timber of value. (Klein and Stroeve 2002 , Hall and Riggs 2007 ). This Forest carbon storage was also calculated for each product is generated for cloud-free snow-covered pixels site using the FACT model, with inputs for each stand using the atmospherically corrected MODIS/Terra sur- based on corresponding FVS-NE simulations of that face refl ectance product (MOD09GHK) and BRDF stand. The FVS- NE model generates estimates of car- look- up tables to correct for surface anisotropy based bon storage through the use of regional allometric on solar zenith angle, sensor zenith, and relative azimuth equations (Jenkins et al. 2003 , Nunery and Keeton together with surface slope, aspect, and MCD12Q1 land 2010 ), and provides information on carbon storage cover type (Klein and Stroeve 2002 ). In MOD10A, within above- and belowground live biomass, the forest snow-covered pixels in non-forested areas are adjusted fl oor in dead and downed wood, and through standing for anisotropic scattering effects using the DIScrete dead wood. Stand simulations in FVS- NE of these Ordinates Radiative Transfer (DISORT) model; snow carbon pools were used to generate relationships optical properties for forested land cover types, as iden- between stand age and carbon storage in order to tifi ed by MCD12Q1, are assumed to be lambertian generate net carbon uptake, a method identical to the refl ectors. MCD43A3 and MOD10A1 both perform well process detailed Gutrich and Howarth (2007 ), but using (biases <0.05) relative to hyperspectral imagery collected FVS- NE simulated carbon tables from each stand over snow- covered forests and cropland mosaics in the instead of Carbon On-Line Estimator (COLE) 1605 state of New Hampshire (Burakowski et al. 2015 ). b reports (reports to the 1605 b program of the United Both snow- free and snow- covered surface albedo States Energy Information Administration, which pro- data were retrieved for each FIA site for the period vide voluntary reporting of greenhouse gas emissions 2002–2013. In order to ensure that each MODIS pixel by public and private entities). We did not model soil was spatially and temporally representative of each carbon, as it is not simulated in FVS- NE owing to forest site, and since precise specifi cs regarding FIA January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 151 plot locations are restricted and fuzzed, we ensured of the stand, we used an exponential decay function that two criteria were met at each location. Firstly, that operated between the non-forest annual mean α each site needed to be located in a MCD12Q1 pixel albedo, c , and the mature forest annual mean albedo, α that was surrounded by pixels of the same land cover m (Cherubini et al. 2012 , Lutz and Howarth 2014 ). type in order to determine spatial homogeneity. The decay function follows the form Secondly, each site needed to remain classifi ed as for- x 𝛼a = ab (7) ested land cover throughout the 2002–2013 MCD12Q1 record. Each site location was paired with the nearest where x represents the stand age and a represents pixel that was consistently classifi ed as cleared (e.g., albedo, and whereupon b is calculated ( ) cropland, grassland, or cropland/natural mosaic) t 0 𝛼 t − through the 2002–2013 record. This pairing was designed b = m m 1 𝛼 (8) in order to calculate canopy-free albedo retrievals and c estimate radiative forcing for each site when harvest was simulated. The mean distance between these pair- and t 0 and t m represent a stand age of zero and the ings was 9.6 km and a fi gure showing the distribution age in which forest canopy saturates and albedo is of these distances can be found in Appendix: Fig. S11. at its lowest. The exponential decay rate for each site Although this approach did not generate albedo data was dependent upon the simulated structural statistics directly for each site, it provided albedo data from from FVS- NE, whereby the time to maturity of the locations generally sharing the same climate and soil stand with respect to albedo was determined to be infl uences, and follows the general approach of Lutz when canopy cover of the main stratum approached and Howarth ( 2014 ). Similarly, we constructed a base- 90% of the maximum cover reached throughout the line mature-forest and a baseline cleared albedo for 200- yr simulation. each site by averaging the mean monthly values as The FACT model generates fl ows of ecosystem ser- collected by the MODIS data over the 11-yr period. vices over a series of annual time steps. Thus, it was necessary to calculate the mean annual radiative fl ux, R , for each site (Bright et al. 2012 ) Net radiative fl ux modeling Ta ∑i=365 In general, we calculated daily shortwave net radia- i−1 RFfs R = . tive fl ux from albedo following the approach of Lutz Ta 365 (9) and Howarth (2014 ) through the use of the FACT model. In this methodology, shortwave radiation fl ux In order to understand the year- to- year net change from a forest stand to the atmosphere is directly related in radiative forcing, RFr, as a result of changes in to the incoming top- of-the- atmosphere solar radiation, forest cover, we then used the equation surface albedo, and two general characteristics of the Δ = R −R RFri Tai Tai−1 (10) atmosphere = R 𝛼 f where i is the year of the simulation. RFfs TOA bsky a (5) in which RF fs is the shortwave radiation fl ux, RTOA Pricing and net present value is the top of the atmosphere solar radiation, calculated We relied on both market and modeled prices to through several equations describing planetary rotation calculate net present values of timber, net carbon (Kalogirou 2009 , Bright et al. 2012 ), α is the sur- bsky uptake, and albedo in the FACT model. Timber values face albedo of the forest stand, and f is a measure a for saw- and poletimber were based on stumpage values of two- way atmospheric transmittance. The calculation reported by the New Hampshire State Department of of f depended upon a measurement of the clearness a Revenue for all three regions of the state from April of the atmosphere, K , at month j, and transmittance T to November 2014 (values available online ). 7 We of the atmosphere, T , through the equation a assumed a price increase at a rate of 1.0% per year f = K T a T,j a (6) based on an analysis by Sendak et al. (2003 ). Each stand was assigned a region of the state depending as used by Bright et al. ( 2012 ). As in Lutz and Howarth upon its geographic position. As the majority of stands ( 2014 ) and Bright et al. ( 2012 ), we used a global mean were composed of more than one species, we calculated of Ta , 0.854, and a clearness index derived from monthly the total stand stumpage value in proportion to the data from the NASA Surface meteorology and Solar two most dominant tree species as determined by total Energy (SSE) product (NASA 2009 ). aboveground biomass throughout the rotation length Forest albedo slowly declines throughout the course of the stand. of the life of a stand as canopy cover increases and the ground surface is masked by vegetation (Kuusinen 7http://www.revenue.nh.gov/mun-prop/property/stumpage- et al. 2014b ). To account for this change over the life values.htm 152 DAVID A. LUTZ ET AL. Ecological Applications Vol. 26, No. 1

Both carbon and albedo shadow prices were calculated where V c( t) represents the marginal benefi t of carbon using the 2007- DICE integrated assessment model sequestration in dollars/ton (1 ton = 1 Mg) as calcu- (Nordhaus 1993 , 2008 , 2010 ). Lutz and Howarth (2014 ) lated in Eq. 12 and Δ C ( t) is the annual incremental describe how this model may be used to assign shadow change in forest stand carbon; and albedo: prices for both carbon and albedo by estimating changes ∑∞ ∏t in social welfare as a result of changes in climate and 1 NPV = V (t) ΔRF (t) . temperature and their associated damages. Social welfare a a r +r (t) t= i=1 1 (16) (W ) in DICE is modeled over decadal time steps, and 0 is based on an instantaneous utility function (U ), which The total net present value of the forest stand is thus is dependent on per capita consumption (c ) and total calculated by adding the net present values of each population (L ), and a discount factor refl ecting the service together relative weight attached to present and future well- being = + + (R ( t ); Nordhaus 2008 ) NPVtot NPVt NPVc NPVa. (17)

T ∑max [ ] W = U c (t) ,L (t) R (t) . Simulations and statistical analysis (11) t=1 We simulated each of the 498 forest sites for a We use this equation to derive the shadow price of total of 1000 yr, long enough to minimize truncation error, using the FACT model to generate net present carbon emissions (V c ) by calculating the change in welfare in response to a one-unit increase in emissions (E ) divided values for each of the three ecosystem services. At by the present- value marginal utility of consumption each site, we simulated all rotation lengths between [ ] 1 and 300 yr, thereby generating a total of 300 sepa- 𝜕W∕𝜕E (t) rate simulations per site. By calculating the total NPV V (t) = [ ] . c 𝜕W∕𝜕c (t) (12) for each rotation length for each site, we were able to estimate the optimal rotation length, whereby the This same technique can be used to calculate the net present value was maximized, using a grid search shadow price of one unit of radiative forcing ( Va ) by technique. In order to determine the infl uence of the simply substituting a unit of radiative forcing (RF) albedo baseline on the optimal rotation length, we for a unit of carbon emissions (E ) calculated two other sets of baselines for each site. [ ] The fi rst additional baseline (Δα ) used the maximum 𝜕W∕𝜕RF (t) max V (t) = [ ] monthly albedos from cleared pixels and the minimum a . 𝜕W∕𝜕c (t) (13) monthly albedos from mature forest sites to calculate cleared and mature baseline albedos wherein the dif- As noted by Nordhaus, the shadow price of carbon ference between the two would be maximized. The Δα is highly sensitive to assumptions concerning future second additional albedo baseline ( min) used the CO 2 emissions trajectories. For the sake of this analysis, minimum monthly albedos from cleared pixels and here we limit attention to the case where emissions the maximum monthly albedos from mature forest are chosen optimally at each point in time to maximize sites to generate baseline albedos where the difference social welfare. See Lutz and Howarth (2014 ) for an between the two would be minimized. Choosing these analysis of the case where policies are chosen to imple- extremes would therefore demonstrate how the selec- ment the 2°C global warming target embodied in the tion of baselines could most drastically alter rotation 2009 Copenhagen Accord. lengths and NPVs. Estimates of the total net present value of the stand We assessed the change in optimal rotation length with respect to all three ecosystem services were cal- for each site as a function of its ecological and bio- culated using the NPV equations for each ecosystem geophysical characteristics (Table 1 ) through the use service: timber of a regression tree approach. Regression trees are

s frequently used in ecological studies to understand the ∑∞ ∏h×i ( ) ( ) 1 variation in a single response variable, in our case NPV = P s ,s ×i V S t i h h t h 1+r (t) the optimal rotation length, associated with other i=1 t=1 (14) explanatory variables (Nair et al. 2013 ). This meth- where P represents stumpage price, s h is stand age at odology uses binary recursive partitioning to generate rotation length h , V is timber volume calculated in splitting rules to successively reduce variability in the Eq. 1, and r (t ) is the discount rate, which is calculated distribution of the response variable across the range endogenously within DICE and ranges from 0.045 to of predictor variables for each site (Moisen and Frescino 0.037, decreasing steadily with time; carbon 2002 ). We chose as the single fi nal tree the largest ∑∞ ∏t for which any additional split would increase the 1 NPV = V (t) ΔC (t) improvement score (as measured by the reduction in c c +r (t) i i=1 1 (15) the sum of squared prediction errors) by more than January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 153

T ABLE 1 . Relative importance of each of 19 predictor variables on optimal rotation length as calculated in the regression tree algorithm.

Variable (unit) Rotation period (%) Albedo (%) Carbon (%) Timber (%)

Latitude (°) 4 1 1 0 Canopy saturation time (yr) 2 0 21 1 Percentage coniferous 2 0 0 1 Elevation (feet) 5 0 0 0 Timber growth rate (%/yr) 28 5 7 57 Maximum carbon storage (tons/ha) 3 0 50 8 Carbon growth rate (%/yr) 19 5 9 31 Sawtimber price ($/m3 ) 7 0 0 0 Poletimber price ($/m3 ) 2 0 0 0 Aspect (°) 2 0 0 0 Slope (°) 2 9 0 0 Classifi ed forest type (%) Broadleaved 0 0 0 0 Mixed forest 1 0 1 0 Grassland 0 6 1 0 Cropland 1 0 0 0 Natural mosaic 1 1 0 0 Saturated albedo baseline 1 1 0 0 Cleared albedo baseline 8 28 3 0 Difference between saturated and cleared 12 44 5 0 baseline

Notes : all prices are given in US$; 1 ton = 1 Mg, 1 foot = 0.31 m. Each variable is listed by relative overall percentage contribution to the prediction of the response variable. Saturated and cleared albedo baseline, and the difference between the two, were all unitless.

1%. To understand the relative infl uence of each site 10 km of each other, thus indicating that specifi c site characteristic on optimal rotation length, we also cal- variables can drastically change optimal strategy of culated a tree-based measure of relative predictor forest harvest. importance for each variable (Breiman et al. 1984 ). The magnitude of the optimal rotation length infl u- Importance of a predictor was calculated as the sum enced the total net present value of the stand. To of improvement scores across all splits in a tree that examine this, we grouped all forest sites into three involved that predictor. We also calculated this relative categories, those with short (<10 yr; mean value importance measure for regression trees in which the $5267.78/ha, SD = $943.28/ha), medium (89–99 yr; net present value of each of the three ecosystem ser- mean value $4338.39/ha, SD = $1261.76/ha), and long vices was used as the response variable. optimal rotation lengths (>200 yr, mean value $2942.00/ ha, SD = $998.39/ha). While there was no statistical difference between the short and medium rotation R ESULTS groups with respect to total NPV, both the short ( t 36 Across all 498 forest sites, the mean optimal rota- = 7.18, P < 0.0001) and medium (t 45 = 4.67, P < tion length when timber, carbon, and albedo were 0.0001) rotation groups differed signifi cantly (P < 0.05) valued was ~94 yr, with a maximum optimal rotation from the long rotation group (Fig. 4 ). The net present of 298 yr, a minimum optimal rotation length of 3 value for each ecosystem services (timber, carbon, and yr, and a standard deviation of ~43 yr. In general, albedo) generally differed depending upon the length optimal rotation lengths shortened from south to north of the optimal rotation length (Fig. 5 ). In sites with across the state (Fig. 2 ). Optimal rotation lengths a very short optimal rotation, albedo was most fre- longer than 200 yr were generally confi ned to the quently the most valuable and har- southern third of the state, while extremely short vesting for timber yielded a substantial cost due to rotation lengths below 10 yr could be found through- the price of mowing. The value of carbon was sub- out the state. A cluster of short optimal rotation stantial for both medium and long rotation lengths; lengths occurred at high elevations near Mount timber value contributed in medium, but not in long Washington within the White Mountain National rotation length sites. Forest. Site- to- site variation differed dramatically; in We examined the infl uence of each of 19 factors several instances, sites with long (>200 yr) and short upon the overall optimal rotation length across all (<10 yr) optimal rotation lengths were located within 498 forest sites using a regression tree analysis. The 154 DAVID A. LUTZ ET AL. Ecological Applications Vol. 26, No. 1

calculated tree (Fig. 6 ) accounts for approximately one-half of the variation in the optimal rotation length across sites ( R2 = 0.51) and has a residual SD of 21.6 yr with a cross- validation SD of 31.9 yr. The branches of the tree reveal that the overall shortest rotation lengths are those sites with a relatively low timber growth rate (<0.01269) and a high difference between the saturated and cleared albedo baselines (>0.1269). These sites have a mean optimal rotation time of only 32.53 yr. Sites with the highest timber volume growth rate have the next shortest mean rota- tion time of 53.68 yr. The longest optimal rotation lengths are associated with very low growth rates and high timber prices. Overall, the variables having the strongest infl uence on the optimal rotation length are the timber volume growth rate, the yearly storage of carbon within bio- mass in the stand (which is closely tied to volume growth), the difference between the saturated and cleared albedo baselines, the cleared albedo value, and the sawtimber stumpage price (Table 1 ). The yearly storage of carbon within biomass and the cleared albedo value do not show up in the tree because they are closely associated with the timber volume growth rate and the difference between the albedo baselines, respectively, which are both stronger predictors of optimal rotation length. In examining the infl uence of F IG. 2 . The optimal rotation length, or period, for each of the each variable on the net present value of each of the 498 simulated sites. Generally, optimal rotation length was shorter three ecosystem services, we fi nd that the albedo dif- in the northern third of New Hampshire, with very short lengths occurring at high elevations in the White Mountain National ference, cleared albedo, and slope were most important Forest (cluster of very short lengths in the eastern upper half of the for the NPV of albedo. For the NPV of carbon stor- state). age, the most infl uential site characteristics are the

1.0

0.9

0.8

0.7

0.6

0.5 Timber, carbon, albedo Timber and carbon 0.4

0.3 Cumulativesites proportion total of

0.2

0.1

0 0 50 100 150 200 250 300 Optimal rotation length (yr)

F IG. 3. A cumulative distribution chart showing optimal rotation lengths for all 498 sites when timber, carbon, and albedo were considered (black line), compared to when only timber and carbon were considered (gray line). Albedo generally shortened optimal rotation length, and in some cases, optimal rotation approached just 3 yr. January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 155

F IG. 4 . A bean plot showing the total NPV for three groups of simulated forest sites. The groups were selected based on optimal rotation length. Each light line represents the NPV for one site, while the darker line represents the mean of the group. The width of the bean displays the distribution of sites within each group. The dotted line indicates the mean net present value of all three ecosystem services across all sites.

F IG. 5. Three bean plots showing the timber, carbon, and albedo NPV for each of the three groups of optimal rotation lengths. For sites with short optimal rotation lengths, albedo and carbon contributed signifi cantly to the total NPV whereas timber harvest, through mowing costs, generated a net negative NPV. Carbon provided the majority of revenues for sites with both medium (middle) and long (right) rotation lengths. Components of plots are as in Fig. 4 .

maximum carbon storage and the canopy saturation The baseline albedo for each site serves as an impor- time. These carbon storage parameters, however, are tant variable in determining the optimal rotation length not especially infl uential in the overall determination for the site. As calculated initially, the baseline for of optimal rotation length. each site was based on mean values for each month 156 DAVID A. LUTZ ET AL. Ecological Applications Vol. 26, No. 1 over the 2002–2013 time period, then averaged across locations, the optimal rotation length approaches zero the years. When we changed the baselines to refl ect (3 years < t < 10 years) at a small percentage of Δα either a max scenario, by choosing the maximum sites. While these do not suggest that continual harvest cleared and minimum saturated albedos for each month, should be prescribed for such locations, given the Δα or a min scenario, by choosing the minimum cleared importance of additional ecosystem services not incor- and maximum saturated albedos for each month, the porated in our analysis such as biodiversity protection mean rotation length and total albedo NPVs were and aesthetic and recreational values, they do suggest signifi cantly different compared with when using the that approaches that limit harvest in an effort to store mean baseline (Table 2 ). The mean rotation length carbon may not provide the greatest bundle of eco- Δα dropped to 59 yr with the max baseline and increased nomic benefi ts from some forest stands. This is of Δα to 112 yr with the min baseline. The albedo net great interest to the burgeoning United States forest present value increased to $2317.27/ha on average for carbon market, as forest owners, including several in Δα Δα the max baseline, whereas for the min baseline, New England, have recently begun to apply for cer- the mean albedo net present value dropped to a nega- tifi cation of carbon offsets through the California Air tive value, at −$208.55/ha. Resources Board offset program. Our results show several locations where a management strategy of a very long rotation length does not provide optimal D ISCUSSION climate benefi ts, suggesting that such strategies may Incorporating surface albedo led to a decrease in be misguided without the inclusion of albedo, a fi nd- optimal rotation lengths for forest sites across the state ing consistent with a growing number of other studies compared to when only timber and carbon were con- (Bright et al. 2014 , Lutz and Howarth 2014 ). sidered (Fig. 2 ). While this general fi nding mimics Furthermore our fi ndings indicate that site- to- site other studies incorporating albedo into an assessment optimal rotation lengths can vary substantially over of optimal rotation lengths (Thompson et al. 2009 , very short distances in temperate forests. While much Lutz and Howarth 2014 ), here we fi nd that in some interest has been placed on albedo with respect to

F IG. 6 . The regression tree used to determine the importance of 19 predictor variables (Table 1 ) on optimal rotation length; n represents the number of sites ultimately partitioned into each grouping. When the stated conditions are true, the tree proceeds to the left, when false to the right. Albedo difference refers to the magnitude of the difference between saturated and cleared baseline albedo for the site. All units are given in Table 1 . January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 157

T ABLE 2. Choice of albedo baseline had a considerable effect on optimal rotation length for the modeled sites NPV (P < 0.0001, n = 498).

Albedo baseline Rotation length (yr) Saturated albedo Cleared albedo Albedo NPV ($/ha)

10- year baseline 94 (44) 0.136 (0.0215) 0.193 (0.0403) 512.76 (823.64) High albedo 59 (46) 0.11 (0.0164) 0.259 (0.0628) 2317.27 (2350.62) Low albedo 112 (49) 0.171 (0.0318) 0.14 (0.0217) −208.55 (243.32)

Notes : Standard deviation (SD) is shown in parentheses. Selecting a reference year where snowfall and albedo are considerably higher generated a nearly 40- yr shortening in rotation length, on average, across all the sites. The opposite effect was true for select- ing a low- albedo reference year, whereby optimal rotations increased on average by nearly 20 yr. forest management in boreal zones with monospecifi c infl uential site variables from the regression tree (timber or low- diversity stands (Bright et al. 2011 , 2014 , growth rate, carbon storage rate, and stumpage price) Kuusinen et al. 2014 a ), New Hampshire ’ s forests often and examined their infl uence on optimal rotation period. contain a wide assortment of species that range from When we reran our simulations with a ±10% change temperate to boreal. Tree species diversity allows for in each variable, we found that the model was most a variety of species- specifi c growth rates and an array sensitive to subtle changes in stumpage price and car- of potential forest products and prices to infl uence bon growth rate (Appendix: Fig. S12). The timber the trade-off of timber, carbon, and albedo ecosystem growth rate did not have a signifi cant infl uence on services. As a result, sites that share similar climates optimal rotation periods, however, at this level of yet differ substantially in species mixes and growth adjustment (10%). When all three variables were rates can yield drastically different mixes of ecosystem adjusted in the same direction, the optimal rotation services and thus different optimal rotation lengths. period across all 498 sites was signifi cantly different Thus, we believe it is critical that forest policies that than in our original runs. Thus, it does not appear seek to maximize climate benefi ts from forests in New that our model is particularly sensitive to subtle vari- Hampshire pay close attention to site-specifi c charac- ations in any of these variables, but that changes to teristics. In other words, generalized models of forest variables that are compounding (for example, timber growth or coarse albedo measurements may not be growth rate and stumpage price) can yield statistically suffi cient for measuring and comparing overall trade- signifi cant differences when compared to a base case. offs of these climate services since such trade-offs are While we did not analyze how the choice of discount easily infl uenced by very slight changes. New Hampshire rate and trajectory may infl uence the model, we hypoth- is situated at a precipitous boundary in which subtle esize that this choice will have a considerable effect. variations in site snowfall or growth rates may change It is important to note that the costs of mowing optimal harvest strategies dramatically. It seems clear precluded the optimal rotation length from reaching that site-specifi c and stand- level research is needed to 1 year and instead, the shortest optimal rotation lengths delineate these boundaries in other states in New were 3–5 years. Short rotation lengths like this can England. in some contexts be used in order to improve habitat Our regression tree analysis of the most important for several bird (Weidman and Litvaitis 2011 ), insect stand variables that infl uenced rotation length led to (Wilson et al. 2014 ), and endangered species such as several insights about site characteristics. The most the New England cottontail ( Sylvilagus tranitionalis ) important variable related to the optimal rotation length (Buffum et al. 2011 ). Additionally, early successional was the rate that forest production was translated into habitat created by this rotation length is also critical timber volume. At forest sites that grew quickly and for understory plant diversity (Swanson et al. 2011 ). generated high-value timber products, such as sites Because of these benefi ts, state wildlife agencies and with pine and maple species, the total duration of federal programs commonly provide fi nancial assistance time that albedo radiative forcing was generated before to pay for the costs of mowing and early timber har- canopy cover lowered surface refl ectivity was short, vest (Oehler 2003 , Buffum et al. 2014 ). In these cases, and thus there was a minimal benefi t toward manag- very short rotation lengths that are subsidized in this ing for albedo (short harvest), compared to longer way may provide even more fi nancial benefi ts to the harvests that maximized timber and carbon storage. manager of the property. In contrast, at sites where timber growth was very The difference in albedo between cleared ground slow, such as high- elevation sites with low- productivity and mature forest was a signifi cant factor in deter- forests, the net benefi ts from timber and carbon took mining the optimal rotation length. These estimates signifi cant time to materialize, and thus, short rotation infl uenced the total magnitude of the radiative forcing lengths that generated albedo benefi ts were optimal. benefi t generated by forest harvest, and therefore the In addition to our regression tree analysis, we per- greater the difference between these two values, the formed a sensitivity analysis based on the three most larger the benefi t obtained by having a short rotation 158 DAVID A. LUTZ ET AL. Ecological Applications Vol. 26, No. 1 length. Previous research in boreal ecosystems has While analyses that incorporate the effect of partial indicated that the difference between cleared and or selective harvest exist regarding trade- offs between mature forest albedo matters greatly in assessing cli- timber and carbon (e.g., Gutrich and Howarth 2007 ), mate trade- offs (Bright et al. 2011 , 2012 ). Our results the infl uence of this type of management on albedo suggest that in regions where snow is intermittent is not completely understood. Kuusinen et al. ( 2014 a ) throughout the winter, the accumulation of snow, and report that a operation did not infl uence hence the wintertime albedo, has a major infl uence shortwave albedo in managed pine stands in Finland, on the rotation length as well. Thus, heavy snow possibly due to the infl uence of the understory. The events and cold winters that foster consistently high presence of standing biomass can drastically impact surface albedo can dramatically change optimal rota- albedo, with even standing dead trees substantially tion strategy compared to warmer, lower snowfall altering measurements taken by satellite (O ’ Halloran winters. et al. 2014 ). Thus, although partial harvests provide While previous research has investigated changes in opportunities for the maintenance of several ecosystem albedo at the beginning and end of the winter season services in managed forest landscapes (Schwenk et al. due to temperature increases (Bright et al. 2014 ), in 2012 ), we do not anticipate that this type of manage- New Hampshire, surfaces do not consistently stay ment will generate albedo benefi ts at a signifi cant scale. snow- covered throughout the winter, and thus indi- Further analyses of the effect of thinning on albedo vidual snow events can alter annual albedo measure- via fi ne- scale imagery in temperate forest stands are ments drastically. Subsequently, we found that choosing necessary moving forward to address the question of a baseline with which to compare historical and future partial harvest in a more complete manner. albedos to calculate net benefi ts is quite infl uential. We did not incorporate the myriad complexities When we altered our assumptions regarding baseline of a changing climate on forest ecosystems into this albedo to represent two different scenarios, we calcu- modeling study. Generally, the New England region lated a substantial change in optimal rotation length of the United States is expected to be warmer and across all forest sites ( P < 0.0001). It is therefore slightly wetter, with increases in annual regional critical that site- specifi c measures of albedo at high surface temperatures between 2.9°C and 5.3°C and temporal and spatial resolution be available to properly an increase in wintertime precipitation of 11–14% assess the appropriateness of climate mitigation projects by 2070–2099 (Hayhoe et al. 2007 ). These changes focused on forests in the state. will undoubtedly infl uence forest species composition Our modeling framework was not capable of address- over the 21st century, with more southerly species ing soil carbon, which can account for nearly 50% of such as oaks and hickories becoming signifi cantly total ecosystem carbon in forests in New Hampshire more dominant by 2100 (Tang et al. 2012 ) and the (Fahey et al. 2005 ). While forest soil carbon is still continued migration of evergreen spruce–fi r forests generally understudied in the context of carbon account- further upslope (Groffman et al. 2012 ). We expect ing (Petrenko and Friedland 2014 ), several authors this to result in markedly different fl ows of timber, have utilized data from chronosequences of forest soils, carbon (Tang et al. 2014 ), and albedo decay rates, and concluded that clear- cut harvest can lead to losses with the overall infl uence being an increased ability of soil organic carbon (SOC), particularly when a large of New Hampshire forest stands to store carbon percentage of biomass is removed (Johnson et al. 2010 ). and generate valuable sawtimber stock. Such shifts Subtle losses in the deep mineral soil may also lead in climate will also likely decrease the total days to overall belowground carbon losses to the atmosphere with snowfall and snow pack (Hayhoe 2007 ), limiting (Vario et al. 2014 ). However, a meta- analysis of 26 the infl uence of albedo and lengthening optimal rota- studies examining soil carbon found that there was tion periods. Our future work to incorporate more little effect of harvest upon the A horizon and no robust forest landscape models (Thompson et al. signifi cant effect on average on B horizon and whole 2011 ) and albedo measurements and models soils (Johnson and Curtis 2001 ). Thus, while there is (Burakowski et al. 2015 ) will help address these cur- no agreed-upon consensus regarding the fate of forest rent uncertainties. soil carbon post- harvest at all levels, it is likely that In conclusion, this research details a modeling heavy harvest under short rotation periods may lead approach to address questions about trade- offs that to losses in the soil carbon pool. Incorporating a loss exist between multiple ecosystem services in temperate of ~10% of the pool (Johnson and Curtis 2001 , Johnson forests in New Hampshire. However, the optimal rota- et al. 2010 ) in our model results would likely lengthen tion lengths calculated and reported in this study should rotation periods slightly, but the short- rotation sites not be interpreted as prescriptive. Practical decisions where albedo is the major contributor would likely regarding optimal forest harvest strategy require bal- be unaffected due to the magnitude of albedo NPV. ancing multiple objectives and preferences and often In any event, this component of the forest carbon involve deciding between a large group of ecosystem budget warrants further attention in FACT modeling services which may or may not have defi ned economic studies moving forward. values. Along those lines, other biophysical infl uences January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 159 on climate from forests such as surface roughness and and albedo dynamics in life cycle impact assessment . surface and latent energy transfer were not included Environmental Impact Assessment Review 37 : 2 – 11 . in our model, although they have been examined else- Bright , R. M. , A. H. Strømman , and G. P. Peters . 2011 . Radiative forcing impacts of boreal forest biofuels: a scenario where (Zhao and Jackson 2014 ). While these properties study for Norway in light of albedo. Environmental Science affect localized surface temperature, their overall impact Technology 45 : 7570 – 7580 . at the global scale is diffi cult to measure (Davin and Buffum , B. , S. R. McWilliams , and P. V. August . 2011 . A de Noblet-Ducoudré 2010 ). Future work will incor- spatial analysis of forest management and its contribution porate the infl uence of cloud albedo on atmospheric to maintaining the extent of shrubland habitat in southern transmissivity, the infl uence of climate change on New England, United States. Forest Ecology and snowfall depth and subsequently snow albedo Management 262 : 1775 – 1785 . Buffum , B. , C. Modisette , and S. R. McWilliams . 2014 . (Burakowski et al. 2015 ), forest successional dynamics, Encouraging family forest owners to create early successional particularly tree species migration, and other valuable wildlife habitat in Southern New England . PLoS ONE ecosystem services such as biodiversity and wildlife 9 : e89972 . habitat. Burakowski , E. A. , S. V. Ollinger , L. Lepine , C. B. Schaaf , Z. Wang , J. E. Dibb , D. Y. Hollinger , J. Kim , A. Erb , and M. Martin . 2015 . Spatial scaling of refl ectance and A CKNOWLEDGMENTS surface albedo over a mixed- use, temperate forest landscape This work was funded primarily through the New Hampshire during snow- covered periods. Remote Sensing of Experimental Program to Stimulate Cooperative Research Environment 158 : 465 – 477 . (EPSCoR) Project “Ecosystems and Society” from the National Canadell , J. G. , and M. R. Raupach . 2008 . Managing forests Science Foundation. Additional funding was obtained through for climate change mitigation. Science 320 : 1456 – 1457 . the Southern Research Station of the U.S. Forest Service Chapin , F. S. , et al. 2000 . Arctic and boreal ecosystems of and the Junior Research Scholarship program at Dartmouth western North America as components of the climate system. College. The authors would like to thank Tom O ’ Halloran Global Change Biology 6 : 211 – 223 . and Georgia Mavrommati for comments on the manuscript. Cherubini , F. , R. M. Bright , and A. H. Strømman . 2012 . D. A. Lutz would like to thank Queso and Waffl e and E. Site- specific global warming potentials of biogenic CO2 A. Burakowski would like to thank Cory for their contribu- for bioenergy: contributions from carbon fluxes and tions to the development of the manuscript. E. A. Burakowski albedo dynamics . Environmental Research Letters would also like to thank the USDA Forest Service Northern 7 : 045902 . Research Station for their assistance relating to this Cherubini , F. , R. M. Bright , and A. H. Strømman . 2013 . manuscript. Global climate impacts of forest bioenergy: what, when and how to measure? Environmental Research Letters 8 : 014049 . L ITERATURE CITED Chladná , Z. 2007 . Determination of optimal rotation period Anderson , R. G. , et al. 2011 . Biophysical considerations in under stochastic wood and carbon prices . Forest Policy for climate protection. Frontiers in Ecology and and Economics 9 : 1031 – 1045 . the Environment 9 : 174 – 182 . Crookston , N. L. , and G. E. Dixon . 2005 . The forest vegetation Bechtold , W. A. , and P. L. Patterson . 2005 . Enhanced forest simulator: a review of its structure, content, and applications. inventory and analysis program—national sampling design Computers and Electronics in Agriculture 49 : 60 – 80 . and estimation procedures. Technical Report SRS-80. USDA Davin , E. L. , and N. de Noblet-Ducoudré . 2010 . Climatic Forest Service, Southern Research Station, Asheville, North impact of global-scale : radiative versus Carolina, USA. nonradiative processes. Journal of Climate 23 : 97 – 112 . Betts , R. A. 2000 . Offset of the potential carbon sink from Dixon , G. 2002 . Essential FVS: a user ’ s guide to the Forest boreal forestation by decreases in surface albedo. Nature Vegetation Simulator . USDA Forest Service , Fort Collins, 408 : 187 – 190 . Colorado, USA . Bonan , G. B. 2008 . Forests and climate change: forcings, Fahey , T. J. , et al. 2005 . The biogeochemistry of carbon at feedbacks, and the climate benefi ts of forests . Science Hubbard Brook . Biogeochemistry 75 : 109 – 176 . 320 : 1444 – 1449 . Fahey , T. J. , P. B. Woodbury , J. J. Battles , C. L. Goodale , Bonan , G. B. , F. S. Chapin , and S. L. Thompson . 1995 . S. P. Hamburg , S. V. Ollinger , and C. W. Woodall . 2010 . Boreal forest and tundra ecosystems as components of the Forest carbon storage: ecology, management, and policy . climate system. Climatic Change 29 : 145 – 167 . Frontiers in Ecology and the Environment 8 : 245 – 252 . Breiman , L. , J. Friedman , C. Stone , and R. Olshen . 1984 . Galik , C. S. , B. C. Murray , S. Mitchell , and P. Cottle . 2014 . Classifi cation and regression trees. WIREs Data Mining Alternative approaches for addressing non-permanence in and Knowledge 1 : 14 – 23 . carbon projects: an application to and Bright , R. M. , C. Antón-Fernández , R. Astrup , F. Cherubini , under the Clean Development Mechanism. M. Kvalevåg , and A. H. Strømman . 2014 . Climate change Mitigation and Adaptation Strategies for Global Change. implications of shifting forest management strategy in a http://dx.doi.org/ 10.1007/s11027- 014- 9573- 4 boreal forest ecosystem of Norway . Global Change Biology Gibbs , H. K. , S. Brown , J. O. Niles , and J. A. Foley . 2007 . 20 : 607 – 621 . Monitoring and estimating tropical forest carbon stocks: Bright , R. M. , R. Astrup , and A. H. Strømman . 2013 . making REDD a reality . Environmental Research Letters Empirical models of monthly and annual albedo in managed 2 : 045023 . boreal forests of interior Norway . Climatic Change Griscom , B. , D. Shoch , B. Stanley , R. Cortez , and N. Virgilio . 120 : 183 – 196 . 2009 . Sensitivity of amounts and distribution of tropical Bright , R. M. , F. Cherubini , and A. H. Strømman . 2012 . forest carbon credits depending on baseline rules. Climate impacts of bioenergy: inclusion of carbon cycle Environmental Science & Policy 12 : 897 – 911 . 160 DAVID A. LUTZ ET AL. Ecological Applications Vol. 26, No. 1

Groffman , P. M. , et al. 2012 . Long- term integrated studies Law , B. E. , and M. E. Harmon . 2011 . Forest sector carbon show complex and surprising effects of climate change in management, measurement and verifi cation, and discussion the northern hardwood forest . BioScience 62 : 1056 – 1066 . of policy related to climate change. Carbon Management Gutrich , J. , and R. B. Howarth . 2007 . Carbon sequestration 2 : 73 – 84 . and the optimal management of New Hampshire timber Le Quéré , C. , et al. 2013 . The global carbon budget 1959–2011 . stands . Ecological Economics 62 : 441 – 450 . Earth System Science Data 5 : 165 – 185 . Hall , D. K. , and G. A. Riggs . 2007 . Accuracy assessment Leak , W. B. 2005 . Effects of small patch cutting on sugar of the MODIS snow products. Hydrological Processes maple regeneration in New Hampshire northern 21 : 1534 – 1547 . hardwoods. Northern Journal of Applied Forestry Hardiman , B. S. , C. M. Gough , A. Halperin , K. L. Hofmeister , 22 ( 1 ): 68 – 70 . L. E. Nave , G. Bohrer , and P. S. Curtis . 2013 . Maintaining Liski , J. , A. Pussinen , K. Pingoud , R. Mäkipää , and T. high rates of carbon storage in old forests: a mechanism Karjalainen . 2011 . Which rotation length is favourable to linking canopy structure to forest function. Forest Ecology carbon sequestration? Canadian Journal of Forest Research and Management 298 : 111 – 119 . 31 ( 11 ): 2004 – 2013 . Hayhoe , K. , et al. 2007 . Past and future changes in climate Lukeš , P. , M. Rautiainen , T. Manninen , P. Stenberg , and and hydrological indicators in the US Northeast . Climate M. Mõttus . 2014 . Geographical gradients in boreal forest Dynamics 28 : 381 – 407 . albedo and structure in Finland. Remote Sensing of Hilt , D. , and R. Teck . 1989 . NE- TWIGS: an individual- tree Environment 152 : 526 – 535 . growth and yield projection system for the northeastern Lukeš , P. , P. Stenberg , and M. Rautiainen . 2013 . Relationship United States . Compiler 7 ( 2 ): 10 – 16 . between forest density and albedo in the boreal zone. Hurteau , M. D. , B. A. Hungate , G. W. Koch , M. P. North , Ecological Modelling 261–262 : 74 – 79 . and G. R. Smith . 2013 . Aligning ecology and markets in Lutz , D. A. , and R. B. Howarth . 2014 . Valuing albedo as the forest carbon cycle. Frontiers in Ecology and the an ecosystem service: implications for forest management . Environment 11 : 37 – 42 . Climatic Change 124 : 53 – 63 . Jackson , R. B. , et al. 2008 . Protecting climate with forests . Malmsheimer , R. W. , J. L. Bowyer , J. S. Fried , E. Gee , R. Environmental Research Letters 3 : 044006 . L. Izlar , R. A. Miner , I. A. Munn , E. Oneil , and W. C. Jenkins , J. C. , D. C. Chojnacky , L. S. Heath , and R. A. Stewart . 2011 . Managing forests because carbon matters: Birdsey . 2003 . National scale biomass estimators for United integrating energy, products, and land management policy. States tree species . Forest Science 49 : 12 – 35 . 109 ( 7S ): S7 – S50 . Johnson , D. W. , and P. S. Curtis . 2001 . Effects of forest McRoberts , R. E. , W. A. Bechtold , P. L. Patterson , C. T. management on soil C and N storage: meta analysis. Forest Scott , and G. A. Reams . 2005 . The enhanced forest inventory Ecology and Management 140 : 227 – 238 . and analysis program of the USDA Forest Service: historical Johnson , K. , F. N. Scatena , and Y. Pan . 2010 . Short- and perspective and announcements of statistical documentation. long- term responses of total soil organic carbon to harvesting Journal of Forestry 3 ( 6 ): 304 – 308 . in a northern hardwood forest. Forest Ecology and Mika , A. M., and W. S. Keeton . 2014 . Net carbon fl uxes Management 259 : 1262 – 1267 . at stand and landscape scales from wood bioenergy Kaipainen , T. , J. Liski , A. Pussinen , and T. Karjalainen . 2004 . harvests in the US Northeast. GCB Bioenergy Managing carbon sinks by changing rotation length in European 7 ( 3 ): 438 – 454 . forests . Environmental Science & Policy 7 : 205 – 219 . Moisen , G. G., and T. S. Frescino . 2002 . Comparing fi ve Kalogirou , S. A. 2009 . Solar energy engineering: processes modelling techniques for predicting forest characteristics . and systems . Academic Press , Burlington, Massachusetts, Ecological Modelling 157 : 209 – 225 . USA . Nair , A. , A. C. Thomas , and M. E. Borsuk . 2013 . Interannual Keith , H. , D. Lindenmayer , B. Mackey , D. Blair , L. Carter , variability in the timing of New England shellfi sh toxicity L. McBurney , S. Okada , and T. Konishi-Nagano . 2014 . and relationships to environmental forcing . Science of the Managing temperate forests for carbon storage: impacts Total Environment 447 : 255 – 66 . of versus on carbon stocks. NASA . 2009 . Surface Meteorology and Solar Energy (SSE) Ecosphere 5 :art75. Release v.6.0 NASA Langly Atmospheric Science Data Center . Kirschbaum , M. U. F. , D. Whitehead , S. M. Dean , P. N. http://power.larc.nasa.gov/cgi-bin/cgiwrap/solar/sse.cgi?+s06 Beets , J. D. Shepherd , and A. G. E. Ausseil . 2011 . Implications Nordhaus , W. 2008 . A question of balance: weighing the of albedo changes following afforestation on the benefi ts options on global warming policies . Yale University Press, of forests as carbon sinks . Biogeosciences 8 : 3687 – 3696 . New Haven, Connecticut, USA . Klein , A. G. , and J. Stroeve . 2002 . Development and validation Nordhaus , W. D. 1993 . Optimal greenhouse- gas reductions of a snow albedo algorithm for the MODIS instrument . and tax policy in the “dice” model. American Economic Annals of Glaciology 34 : 8 . Review 83 : 313 – 317 . Kuusinen , N. , P. Lukeš , P. Stenberg , J. Levula , E. Nikinmaa , Nordhaus , W. D. 2010 . Economic aspects of global warming and F. Berninger . 2014a . Measured and modelled albedos in a post-Copenhagen environment. Proceedings of the in Finnish boreal forest stands of different species, structure National Academy of Sciences USA 107 : 11721 – 11726 . and understory . Ecological Modelling 284 : 10 – 18 . Nunery , J. S. , and W. S. Keeton . 2010 . Forest carbon storage Kuusinen , N. , E. Tomppo , Y. Shuai , and F. Berninger . 2014b . in the northeastern United States: net effects of harvesting Effects of forest age on albedo in boreal forests estimated frequency, post-harvest retention, and wood products. Forest from MODIS and Landsat albedo retrievals. Remote Sensing Ecology and Management 259 : 1363 – 1375 . of Environment 145 : 145 – 153 . Oehler , J. D. 2003 . State efforts to promote early- successional Lamson , N. I. , and W. B. Leak , . 2000 . Guidelines for habitats on public and private lands in the northeastern applying group selection harvesting . USDA Forest Service United States. Forest Ecology and Management Northeastern Area, Newtown Square, Pennsylvania, USA. 185 : 169 – 177 . http://na.fs.fed.us/stewardship/pubs/guidelines/guidelines. O ’ Halloran , T. L. , S. A. Acker , V. M. Joerger , J. Kertis , htm and B. E. Law . 2014 . Postfi re infl uences of attrition January 2016 TRADE-OFFS BETWEEN FOREST SERVICES 161

on albedo and radiative forcing . Geophysical Research Tang , G. , B. Beckage , and B. Smith . 2014 . Potential future Letters 41 : 9135 – 9142 . dynamics of carbon fl uxes and pools in New England Otto , J. , et al. 2014 . Forest summer albedo is sensitive to forests and their climatic sensitivities: a model-based study. species and thinning: how should we account for this in Global Biogeochemical Cycles 28 : 286 – 299 . Earth system models? Biogeosciences 11 : 2411 – 2427 . Thompson , J. R. , D. R. Foster , R. Scheller , and D. Kittredge . Petrenko , C. L. , and A. J. Friedland . 2014 . Mineral soil 2011 . The infl uence of land use and climate change on carbon pool responses to forest clearing in northeastern forest biomass and composition in Massachusetts, USA . hardwood forests. GCB Bioenergy . doi: 10.1111/gcbb.12221 . Ecological Applications 21 : 2425 – 2444 . Price , C. , and R. Willis . 2011 . The multiple effects of carbon Thompson , M. P. , D. Adams , and J. Sessions . 2009 . Radiative values on optimal rotation . Journal of Forest Economics forcing and the optimal rotation age. Ecological Economics 17 : 298 – 306 . 68 : 2713 – 2720 . Ray , D. G., M. R. Saunders , and R. S. Seymour . 2009 . Recent Tittmann , P. , and S. Yeh . 2013 . A framework for assessing changes to the northeast variant of the forest vegetation the life cycle greenhouse gas benefi ts of forest bioenergy simulator and some basic strategies for improving model and biofuel in an era of forest carbon management . Journal outputs. Northern Journal of Applied Forestry 26 ( 1 ): 31 – 34 . of Sustainable Forestry 32 : 108 – 129 . Russell , M. B. , A. R. Weiskittel , and J. A. Kershaw . 2013 . van Kooten , G. C. , C. S. Binkley , and G. Delcourt . 1995 . Benchmarking and calibration of forest vegetation simulator Effect of carbon taxes and subsidies on optimal forest individual tree attribute predictions across the northeastern rotation age and supply of carbon services. American Journal United States. Northern Journal of Applied Forestry of Agricultural Economics 77 : 365 . 30 : 75 – 84 . Vario , C. L. , R. A. Neurath , and A. J. Friedland . 2014 . Schaaf , C. B. , et al. 2002 . First operational BRDF, albedo Response of mineral soil carbon to clear-cutting in a nadir refl ectance products from MODIS . Remote Sensing northern hardwood forest . Soil Science Society of America of Environment 83 : 135 – 148 . Journal 78 : 309 . Schaaf , C. B. , J. Liu , F. Gao , and A. H. Strahler . 2011 . Weidman , T. , and J. A. Litvaitis . 2011 . Are small habitat Aqua and Terra MODIS albedo and refl ectance anisotropy patches useful for grassland bird conservation? Northeastern products. Pages 549 – 561 in B. Ramachandran , C. O. Justice , Naturalist 18 : 207 – 216 . and M. J. Abrams , editors. Land remote sensing and global Williams , C. A. , M. K. Vanderhoof , M. Khomik , and B. environmental change. Springer , New York, New York, Ghimire . 2014 . Post- clearcut dynamics of carbon, water USA . and energy exchanges in a midlatitude temperate, deciduous Schwenk , W. S. , T. M. Donovan , W. S. Keeton , and J. S. broadleaf forest environment. Global Change Biology Nunery . 2012 . Carbon storage, timber production, and 20 : 992 – 1007 . biodiversity: comparing ecosystem services with multi-criteria Wilson , M. K., W. H. Lowe , and K. H. Nislow . 2014 . Family decision analysis . Ecological Applications 22 : 1612 – 1627 . richness and biomass of understory invertebrates in early Sendak , P. E. , R. C. Abt , and R. J. Turner . 2003 . Timber and late successional habitats of northern New Hampshire . supply projections for northern New England and New Journal of Forestry 112 : 337 – 345 . York: integrating a market perspective . Northern Journal Wykoff , W. R. , N. L. Crookston , and A. R. Stage . 1982 . of Applied Forestry 20 : 175 – 185 . User ’ s guide to the stand prognosis model . Technical Report Sturm , M. 2005 . Changing snow and shrub conditions affect INT-22. USDA Forest Service Intermountain Forest and albedo with global implications . Journal of Geophysical Range Experiment Station, Ogden, Utah, USA. Research 110 : G01004 . Yaussy , D. A. 2000 . Comparison of an empirical forest growth Swanson , M. E. , J. F. Franklin , R. L. Beschta , C. M. and yield simulator and a forest gap simulator using actual Crisafulli , D. A. DellaSala , R. L. Hutto , D. B. Lindenmayer , 30- year growth from two even- aged forests in Kentucky. and F. J. Swanson . 2011 . The forgotten stage of forest Forest Ecology and Management 126 : 385 – 398 . succession: early- successional ecosystems on forest sites . Zhao , K. , and R. B. Jackson . 2014 . Biophysical forcings of Frontiers in Ecology and the Environment 9 : 117 – 125 . land- use changes from potential forestry activities in North Tang , G. , B. Beckage , and B. Smith . 2012 . The potential America . Ecological Monographs 84 : 329 – 353 . transient dynamics of forests in New England under Zhu , K. , C. W. Woodall , and J. S. Clark . 2012 . Failure to historical and projected future climate change . Climatic migrate: lack of tree range expansion in response to climate Change 114 : 357 – 377 . change . Global Change Biology 18 : 1042 – 1052 .

S UPPORTING INFORMATION Additional supporting information may be found in the online version of this article at http://onlinelibrary.wiley.com/ doi/10.1890/14-2207.1/suppinfo

D ATA AVAILABILITY Data associated with this paper have been deposited in Dryad: http://dx.doi.org/10.5061/dryad.5vv6t