In the absence of a national policy to lower net GHG emissions, deliberate policies in the US have instead been enacted at the state level. State energy policy makers and their constituents can better understand the range and scope of local emissions sources which feed climate warming trends through periodic greenhouse gas (GHG) accounting. Measuring emissions levels also establishes the baseline by which states can benchmark mandatory reduction targets. Eighteen US states have passed global warming reporting requirements to track state-based contributions to climate change, with three more state reporting mandates in progress [1] . As example, the Connecticut General Assembly’s 2008 passage of Public Act 08-98, an Act Concerning Connecticut Global Warming Solutions (GWSA), introduced a triennial GHG inventory reporting requirement for that state [2] . The complexity of information behind emissions data collection, as well as the need for standardization of collection protocols among states for comparison at the federal level, has given rise to default data aggregators such as the US Environmental Protection Agency’s (EPA) State Inventory Tool (SIT) [3] . The EPA SIT is the principal data collection tool for assembling state-level GHG inventories using data collected by US federal agencies and default estimates of C content and combustion efficiencies. EPA bases the land use, change and forest (LUCF) module of the SIT on GHG guidelines developed for country-wide assessment of land-based C sinks [4] . This approach requires a consistent, comprehensive division of land-based biomass according to the six main IPCC land use categories―forest land, cropland, grassland, settlements, wetlands and other lands―but does not provide single states the measurement tool to assess ground-level changes in biomass that alter C accounting results over time.

Many public agencies and research institutions have developed proprietary methodologies for assessing the impacts of GHG emissions on the atmosphere, some of which include methods to account for LUCF C sinks [5] - [7] . Assessment tools have varied, as have the scopes of application, results and conclusions. CT utilizes the EPA SIT, with its “top-down” approach to GHG accounting, except in the LUCF sector since the SIT LUCF module has not yielded consistent land use accounting results [8] . Assumed changes in LUCF accounting methodology reported forest C flux divergent in the extreme and which converted the sector from C sink to source between 1997 and 1999¹. Reporting uncertainty has resulted in the LUCF sector’s omission from subsequent state inventory protocols until better methodologies could be identified.

Sohl et al. [9] call for a national, consistent land use methodology to measure C flux, offering a prototype which demonstrates LUCF change based on biogeochemical modeling. The US Geological Survey in turn establishes emissions profiles for four specific US eco-regions using the FORE-SCE model to spatially map downscaled scenarios which would contribute to a national assessment of C stocks [10] . However, neither of these novel approaches to land-use emissions accounting is user-friendly at the level of state GHG inventory assembly.

The availability of cumulative land use data in CT over a twenty-five year span elucidates how long-term loss of biomass through sprawling land conversion compromises urgent policy goals of slowing global warming. Mapping data gathered at five-year periodic intervals provides consistent opportunity to analyze transitional land change dynamics. The land cover classes most heavily impacted by land conversion are the state’s terrestrial carbon (C) sinks―most prominently forests, but also green fields, soils, and above- and below-ground biomass. While the environmental role of these C sinks to climate regulation is well understood, methods to quantify their contribution to regional C budgets to date have largely been ignored due to the absence of reliable C accounting. This study seeks to address the weakness surrounding C sink accounting by proposing a ground-up methodology to quantify land-based C sequestration which can used by climate-oriented policy makers.

Research on emissions increases from development-related land cover change supports the proposition that LUCF should play a greater role in GHG reduction strategies [11] - [14] . From an ecological perspective, land converted from both forests and farmland implies loss of wildlife habitat, habitat fragmentation, deterioration of air and water quality, groundwater run-off from fertilizers and pesticide, impeded groundwater recharge with the expansion of impervious surfaces, and disappearance of farmland. From a global warming perspective, land use characterized by low development density impacts the region’s GHG profile, raising emissions from lengthened car travel, deterring public transit use, and increasing the minimum zoning requirements for property and, often in consequence, the C footprint of individual households. From a social equity standpoint, the continued trend toward outlier development diverts state and federal monies away from existing infrastructure improvement in older, inner-core communities in need of reinvestment and toward peripheral development, fueling the cycle of sprawl through newly-lain access to low-density areas [15] - [17] . Satellite land mapping supplies grounds to conclude that indiscriminate land use both drives GHG increases from mobile sources and exacerbates these dynamics by reducing the capacity for terrestrial C storage within the same landscape [18] .

New England states have seen a downturn in secondary forest recovery since 1970 [19] . CT has witnessed an increasing amount of undeveloped forest land being converted to development from 1985 to 2010 [20] . Land development encompasses not just residential and commercial building as well as roads, subdivisions and other incursive forms of urban and suburban land use. Land conversion in CT has been piecemeal, incremental and often decentralized at the level of township or property owner, masking the long-term regional impact of development and deforestation. Forest conversion to low-density housing (6 - 25 homes/km²) was considered the fastest growing driver of New England’s land cover change over this period [21] . In-state population growth is a key indicator value when gauging the efficiency of Connecticut’s residential land development. US Census data reveals a highly skewed growth ratio of population growth in CT to developed land use, using a low-threshold housing development density of at least one housing unit per four acres to plot land development by census tract. In percentage terms, rates of land development grew at roughly eight times the population increase over the period 1970-2000 [21] (Figure 1). As a point of contrast, the Pacific Coastal Region forecasts forest cover reduction of four percent for the fifty-year period of 1997 to 2050 [22] . Over the next decades, CT’s ratio of forest lost to population growth is expected to exceed Washington’s by six-to-one, without accounting for future demographic shifts west.

CT lies in the wet, temperate climate zone of southern New England at 40˚58'N to 42˚3'N and 71˚47'W to 73˚44'W. Elevation varies from sea level along the southern coast of Long Island Sound to 2316' (706 m) in the northwest corner adjoining New York and Massachusetts. CT is known for its attractive landscapes of densely forested hills. Forestlands cover 1.76 million acres (2742 miles²) [20] , 80% of which are deciduous forests. Deciduous forests represent the largest C stock in CT, hovering just below one-half of the combined total for all land use classes in 2010. Twenty percent of CT’s forest land is classified as urban forests located near or within urban zones; half (49%) of these forests lie within 100 yards of hard development or agriculture [23] . New London

County abutting CT’s southern border is forecast to lose between 40% - 63% of private forestlands to land conversion and development by 2031, a prediction similar for New England’s mid-coastal region [19] . Heavy development, rapid forest conversion and increasing sprawl extending from aging commercial arteries will continue to endanger Southern New England’s coastal forests. Changing land cover in CT over the period 1985-2010 shows clear growth in developed areas (red) concentrated along the CT River valley dividing the state as well as its southern coastline (Figure 2).

The land use component of a state GHG inventory ideally reflects localized data gathered through satellite mapping similar to the type used under the REDD mechanism (Reducing Emissions from Deforestation in Developing Countries) to document avoided deforestation. To employ the REDD approach locally, satellite mapping data of state-wide land cover would be necessary to facilitate a “bottom-up” alternative to the SIT LUCF module. The Center for Land Use Change and Research (CLEAR) at the University of Connecticut (UConn) had designed a geospatial mapping methodology for NASA through the use of satellite data to monitor land change over time dating from the early 1980s [20] . The results of this long-term mapping project provided the landscape-scale data set to model changes to the state’s terrestrial C sinks. CLEAR’s satellite mapping model demonstrates the effects of development on CT’s wooded landscape. The diagnostic value of CLEAR’s satellite mapping project is widely accredited in documenting the imprint of urban sprawl [24] - [27] but had not previously been applied to estimate changes to terrestrial C sinks in CT.

The InVEST Carbon Storage and Sequestration computer model, developed at Stanford University’s Wood Institute for the Environment and the Natural Capital Project [28] provided the simulation software to quantify and track terrestrial C storage in CT. The InVEST Version 2.5.3 model uses a raster data set drawn from land cover maps of the type produced by Project CLEAR in an ARC-MAP format [29] . A second required data set for modeling changes to sequestered C is an estimate of C pool valuations for each of the LUCF classes, assembled from the scientific literature on C density and terrestrial C capture specific to Southern New England forest eco-systems.

Nearly five percent of undeveloped land in Connecticut (CT), the subject of this study, and close to four percent of its forests, was converted to development from 1985 to 2010. The conversion of undeveloped land categories to development, turf, other grasses, and barren land totaled 5.12% at the end of the twenty-five year study period. However, the percentage reduction in C sequestration significantly exceeded the 5.12% rate of land cover change due to the shift from high-density C biomass to low- or no-density C land class cover. Land converted from classes of higher C density (deciduous forests: −3.30%, coniferous forests: −0.32%; forested wetlands: −0.19, and agricultural fields: −1.27) to categories of lower C density (developed: +3.00%; turf grass, including residential lawns: +1.51%; other grasses, e.g., highway borders: +0.61%; and barren land: +0.19%) lowered the state’s aggregate C stock responsible for sequestration.

A static modeling of C sequestration, assuming no biomass growth within each vegetation type, yielded a net loss of C sequestration of 38.23 MMTCO₂, an amount equal to a 3.7% sequestered C reduction from 1985 to 2010. Loss of sequestering capacity was led by a 3.3% reduction in deciduous forest cover. Reduced C closely approximated the 3.83% forestland converted over the same timeframe. In 1985, forested land classes contributed 84% of total sequestered C, with deciduous forests providing fully 73% of total sequestration. By 2010, C sequestered by deciduous hardwoods had declined by 2% relative to total C sequestered by all land cover classes; however, total C sequestration shrank 22.4% from 1985 to 2010, reflecting the impact on C capture of reduced forest cover.

However, New England’s forests are successional and increasing in biomass. For these dynamic biomass growth scenarios, we assumed forest biomass increased by an annual mean growth rate of 2.5 MgC/ha⁻¹/y⁻¹ for deciduous forests and 3.0 MgC/ha⁻¹/y⁻¹ for coniferous forests, and thus the rate of potential C sequestration was compounded across time for existing forests. Growth rate scenarios of ±10% and ±40% of baseline value captured any significant over- or under-estimation of C stock values for the two principal forest classes―deci- duous and coniferous―which essentially determined the outcomes of the InVEST C sequestration simulations. Running scenarios at values for the isolated forest classes conveyed the degree of influence of cumulative biomass stock growth values on overall sequestration goals. Figures 3(a)-(d) depict the results of these various sensitivity analyses run with the land conversion data provided by CLEAR.

Sequestration scenarios held constant all baseline values of C density per non-forested land cover type while adding a growth model for either or both forest types. The undulation in the early curves showed a decline in C capture owing to an atypically larger loss of forest cover from 1990 to 1995. This dip indicated that the degree of C sequestration from natural biomass growth was not sufficiently robust to overcome the loss of C density due to the higher rate of deforestation during the period. Deforestation in the final C inventory years was slightly slower and, combined with a more robust volume of vegetative growth in this latter period, yielded stronger levels of C sequestration in the final phase of the confidence interval scenarios.

Modeling of all scenarios revealed the ±10% variation from baseline of combined deciduous and coniferous C stock to be the prime determinant of long-term C sequestration in Southern New England forests. The slopes of these four sequestration trajectories and their confluence at the 2010 data interval suggested that C stock values were only overtaken by the most extreme variations from baseline values of ±40% deciduous and coniferous biomass growth to decide levels of sequestration. The highest overall level of C sequestered occurred in the scenario describing 40% above baseline biomass growth by both deciduous and coniferous forests (+40% D & C biomass growth) to yield 1214.5 MMTCO₂. The +10% upper limit C stock for deciduous and coniferous forests (+10% D & C C stock) came in nearly at par at 1212 MMTCO₂―only 2.5 MMTCO₂∙ha⁻¹∙y⁻¹ short of the total sequestration for the +40% D & C biomass growth scenario. The +10% deciduous biomass growth scenario at 1198 MMTCO₂ ranked third in terms of total sequestration, trailed closely by +40% deciduous only biomass growth scenario at 1196 MMTCO₂.

Based on our sensitivity analyses of variations from baseline values across both C stock and biomass growth rates, total C sequestered in CT, in CO₂ equivalents fluctuated most widely in response to increases or decreases in underlying C biomass stock. A ±10% variation from baseline value of combined D & C forest stock yields a 47.8 MMTCO₂ increase or decrease in total C sequestration [(1212 − 1019)/2), whereas a ±10% variation in baseline value of deciduous and coniferous annual biomass growth produced a 24.5 MMTCO₂ difference from baseline C capture [(1214.5 − 1017.7)/2]. C stocks proved nearly twice as influential in determining total C sequestration across the twenty-five years.

Assumptions of accumulating C stocks quantitatively pushed total C sequestration from a 1985 baseline level of 866 MMTCO₂∙ha⁻¹∙y⁻¹ to a 2010 baseline value which reached 1116 MMTCO₂∙ha⁻¹∙y⁻¹, or an increment of 250 MMTCO₂∙ha⁻¹∙y⁻¹. The upper and lower limits of ±10% of baseline value of forest C stock show fairly predictable levels of C sequestration for the period 1985-2010. Deciduous C stock is the most variable. Sequestered C for the lower limit of -10% baseline C stock for deciduous and coniferous forests grew by 226 MMTCO₂ over the 25-year study period, as compared to 252.8 MMTCO₂ for the +10% upper limit of baseline C stock for deciduous and coniferous forests. A 10% over- or under-estimation of C stocks produced a difference of 13.5 MMTCO₂ from the 2010 baseline, which equates to 35% of CT’s 2010 CO₂ emissions level (38.12 MMTCO₂). This result emphasizes the responsiveness of sequestration processes to the underlying change in forest C stocks.

Modeling results showed that the percentage of foregone sequestration estimated for the baseline model was 17.68%, with modeling extremes ranging from 14.59% to 23.82%, corresponding to ±40% deciduous and coniferous biomass growth scenarios, respectively. The highest level of foregone C sequestration was estimated at over 60 MMTCO₂ for a +10% over baseline of forest C stocks. Even the most modest measure of foregone C sequestration (47.34 MMCTO₂ from −10% variation to baseline C stock value) in excess of baseline sequestered C values surpasses accounting of CT’s 2010 total fossil fuel emissions (38.12 MMTCO₂), expressed as CO₂ equivalents.

2010 foregone sequestration expresses the difference in MMTCO₂ between realized C sequestered over 1985-2010 and the amount of foregone sequestered C due to deforestation by the end of the study period. While the total twenty-five year gain in sequestered C for the baseline scenario was 250 MMTCO₂, the baseline scenario for foregone C sequestration in CT during 1985-2010, given 1985 land use patterns, is the C equivalent of 303.9 MMTCO₂. This baseline scenario values indicate a loss of 53.74 MMTCO₂, the equivalent of 17.68% of potential C sequester for the full twenty-five years as determined by subtracting 1985 estimate of total sequestered C from the 2010 estimate.

Comparing the modeling results of C sequestration summarized in Figure 4 steers attention to the significant difference between the percentage of forest cover loss (3.83%) and the percentage of foregone C sequestration (17.68%) occurring over this period. The divergence between these percentages owed to the higher C density of deforested biomass relative to the land cover classes which replaced it. The 3.83% forest cover reduction represented 190.1 square miles and comprised most of the 5.12% land use conversion in the categories of developed land, turfgrass and other grasses. The 17.68% baseline C sequestration reduction rate over the 1985 baseline scenario was 4.62 times greater than the 3.83% baseline forest conversion rate. Sensitivity analyses of the upper and lower limits to these baseline values reinforce these comparative results to only slightly greater and lesser degrees.

The contrast of twenty-five year history of CT fossil fuel emissions derived from the CT Emissions Summary,

1990-2010 of the EPA State Inventory Tool [3] , with foregone opportunities for C sequestration which vegetative land converted to lower C density failed to capture, is striking. The 2010 fossil fuel emissions total was calculated from data available at the Energy Information Agency [60] . CT’s level of annual CO₂ emissions has remained fairly stable slightly above or below the 40 MMTCO₂ mark, with some vicissitudes reflecting state economic conditions and recent policy mandates to curb emissions in light of growing awareness of climate change. The use of 1985 as baseline year for measuring foregone C sequestration corresponds to the start date of CLEAR’s land use change inventorying and was thus recorded as zero. Data comparisons showed that the additive loss of sequestered C, significant in its own right, rose over time due to the continued trend in biomass growth, with a surging gap between emissions and foregone sequestration expected in future years.

Comparing sequestered C against loss of sequestered C due to recent deforestation revealed the degree of impact CT’s land use classes have on state’s efforts to mitigate global warming. Modeling of sequestered C from existing biomass growth without land use change rose from 0 in 1985 to 53.74 MMTCO₂ in 2010. Under this scenario, the annual level of foregone sequestered C would have surpassed CO₂ emissions around the year 2002 and by 2010 would have equated to approximately 140% of emissions measured in 2010 by CT’s most recent GHG inventory [61] (Figure 5).

The expanse of deciduous forests in Southern New England states like CT, amplified by the high level of C they sequester relative to other land uses in the state, positions them as the most impactful land class for our modeling of sequestered C. The high sequestration level of above-ground deciduous biomass (AGB) of baseline value, estimated at 109.8 MgC∙ha⁻¹∙yr⁻¹ for 1985 [23] but reaching 172.3 MgC∙ha⁻¹∙yr⁻¹ by 2010 according to the dynamic growth modeling, speaks to the influence of these stands as a stabilizer for local C budgets.

The natural processes of forest succession will continue to spur increases in rates of C captured by older forest stands due to changing forest species composition with time. With the exception of natural disturbances, longer- lived, shade-tolerant mid- to late-successional species will replace earlier successional species, resulting in increased C sequestration. This owes to the fact that late successional species absorb more C due to the higher biomass density in their woody parts. Research on New England forests shows that mid- to late-successional tree species have higher levels of C uptake than their predecessor species [62] . The amount of stored C could more than double in protected forests as they reach maturity [63] .

Land conversion and climate change hold the potentially greatest impact to forest landscapes globally [30] . Successional forest growth and regional recovery following a relatively recent history of agricultural abandonment should prevail in driving forest dynamics over the next half century [64] . C sequestration by forest biomass is identified as one of a few climate-related gains to function as a negative feedback loop by buffering continued warming trends [30] . The percentage of modeled sequestration loss in CT over the twenty-five year study period was 4.62 times the percentage loss of high C-density forestland for the same time span. Figure 6(a) and Figure 6(b) express normalized rates of changes for related land cover categories, indicating the high sensitivity of forest biomass contraction to C sink value. Continued relative land pattern changes as shown here will slow the negative feedback loop C sinks provide and, in contradiction, hasten the impacts of climate change.

Periods of active C sequestration by New England’s forest stands have been shown to extend decades beyond previous research assumptions [64] . Despite a century-old legacy of natural forest recovery, Southern New England will continue to see forest C sequestration for decades into the future. Land conservators, whether public or private agents, must take into account that the equilibrium point for C storage has not yet been reached. Forest growth and succession will potentially have the largest impact on the aboveground C stock, and this study has shown that deciduous forests in particular have a disproportionate influence on CT’s emissions future. Preventative forest conservation ought therefore to play a key role in any global warming solution policy.

Research which identifies land use mismanagement as the enabler of suburban sprawl also proposes the use of satellite land use mapping for better natural resource management [65] . The ecological relevance of mapping for public policy-making is value-additive, embracing habitat conservation, climate adaptation, and water quality management into the process of urban development and regional transportation planning for whole-design land use planning.

Comparing sequestered C against loss of sequestered C due to recent deforestation revealed the degree of impact CT’s land use classes have on state’s efforts to mitigate global warming. In order to capture GHG remediation through C sink preservation, low-density development in areas of mature, second-growth forests should be discouraged, with development diverted toward renewing the original suburban tracts ringing the urban cores, first built in the 1950s but now suffering from neglect, aging infrastructure and blight. The logic of redirecting development inward holds many merits from a sustainability perspective, among them concentrating population density for transit viability, reducing costs of installation and maintenance for roads and utilities, shortening commuting time and distance, and lowering transportation emissions. Fiscal incentives which reward diverted development can be dually aimed at avoided deforestation, a mechanism largely absent from current climate policy and which itself has not been monetized in terms of economic and social benefits at local and regional planning levels. Avoided deforestation is a major lever for C abatement and should be considered the equivalent of local negative feedback in the way local emissions contribute positively to climate change.

This study considered land use change in fine-grain detail to assess the environmental effects of state-wide, incremental land conversion, principally of deciduous forestlands which comprised high-density C storage pools. The CLEAR satellite mapping of CT performed over the last two and a half decades provided cumulative data on LUCF and land cover conversion to new analytical tools adapted to model responses to climate change. CLEAR’s land cover data set is to our knowledge the only state-wide mapping project in the United States of this longevity [66] . The application of this data for the analysis presented here demonstrates novel possibilities for quantifying and valuating C removal in the light of LUCF analysis―a process heretofore disregarded for accounting reasons―to better inform environmental policy choices that come before state and regional governments.

The results of this research prove the importance of land use in achieving a full spectrum of C accounting which includes both CO₂ emissions output and C sink estimation. Estimates of C sequestered within all land use categories and the changes within each type create a three-dimensional mosaic of land utility for states to construct global warming solutions. Energy, transportation and environmental policy-makers can benefit from the analysis of land use mapping in order to integrate carbon sink preservation into state policy-making aimed at combating climate change.

We wish to acknowledge the expertise at the institutions which were instrumental for setting this study in motion: the CT Department of Energy and Environmental Protection (DEEP), specifically DEEP Commissioner Dan Esty, Bureau Chief Tracy Babbidge, and climate policy experts Lynn Stoddard and Pete Brunelli; Chet Arnold and Emily Wilson for their pioneering work at CLEAR; Bradley Eichelberger at Woods Institute for the Environment at Stanford University; the staff at the Center for Geographic Analysis at Harvard University; and Jacob Sobstyl, doctoral candidate at MIT, for assistance in modeling our results.