Delineation of Roadless Areas in the Northern Forest of New England Using Satellite Imagery

Delineation of Roadless Areas in the Northern Forest of New England Using Satellite Imagery David Publicover Cathy Poppenwimer Appalachian Mountain Club Research Department P.O. Box 298, Gorham, NH 03581 AMC Technical Report 02-1 February 2002

Executive Summary In 1999 AMC researchers undertook a study to identify roadless areas greater than 2,000 hectares (4,940 acres) in size across a 9.4 million hectare (23 million acre) region of northern Maine, New Hampshire and Vermont. The purpose of the study was to assist AMC and other interested partners by identifying potential conservation priorities and specific conservation proposals. Today roadless areas are one of the most ecologically significant components of a landscape, especially for the identification of areas suitable for designation as ecological reserves, wilderness or other natural areas. Roadless areas are the most likely parts of a landscape to contain large expanses of mature unfragmented forest habitat of high ecological integrity. Large roadless areas are relatively scarce in northern New England due to the long period of human use, the relative lack of large public lands, and the presence of extensive large commercial forest management ownerships. Because existing digital road data does not accurately represent the extensive and ever-changing network of private roads across the region, our primary source of information was recent (1996-98) satellite imagery (Landsat Thematic Mapper and Indian Remote Sensing). This imagery clearly showed most roads and recent heavily harvested areas. Other sources of information were up-to-date Delorme road atlases and historical forest clearing data developed by AMC in a previous study (which showed areas of forest that had been cleared at any time between the late 1960s and 1991). Roadless areas at least 2,000 hectares (4,940 acres) were digitized on-screen over the satellite imagery following bounding roads and heavy harvest areas according to the following criteria: 1) no roads evident on the recent satellite imagery, 2) no improved roads shown on Delorme atlas maps, and 3) no evidence of heavy harvesting visible on either the recent satellite imagery or identified in the historical forest clearing study. The areas identified in this study are not pristine primary forest. All have been harvested in the past, and many contain unimproved, abandoned, or otherwise low-impact roads and logging trails. Areas may have undergone less intensive recent harvesting not evident on the satellite imagery. Thus it is more appropriate to characterize these areas as very low road density rather than truly roadless. However, they clearly represent the least heavily impacted parts of an otherwise heavily roaded and harvested landscape. The study identified 137 areas ranging in size from 2,000 to 78,000 hectares (4,940 to 193,000 acres). In total they encompass over 990,000 hectares, or approximately 11% of the total study area. 78% of the areas were less than 8,000 hectares in size, with a median of 4,300 hectares. The largest area encompasses much of Baxter State Park in Maine and adjacent lands. Roadless areas are disproportionately concentrated on public conservation lands. Public lands make up 10% of the study area but encompass 44% of the total roadless area. About 49% of total public land area but only 6.5% of private land was mapped as roadless. Though there was no overall correlation between the size of a roadless area and the proportion of that area in public

ownership, there is a relationship between large roadless areas and large blocks of public land. The 15 largest blocks (those above 15,000 hectares in size) are all located at least partially on a large block of public land. These large blocks include 7 in the White Mountain/Mahoosuc Range region, 3 in the Baxter State Park region and 2 in the Mount Mansfield/Camel s Hump region. There is also a strong relationship between elevation and the location of roadless areas. 12% of roadless area acreage is above 2700 in elevation, though these high-elevation lands comprise less than 2% of the study area. Lands above 1500 make up 16% of the study area but encompass about half of the roadless areas. Less than 5% of land below 1700 was mapped as roadless, compared with over 70% of land above 2700 and almost all land above 3700. This study represents a snapshot in time of an ever-changing landscape. Since the decline of river drives and logging railroads and the rise of road-based transportation, an ever-expanding network of logging roads has essentially become a permanent feature of the landscape. Many of the areas identified in this study have undergone road construction and harvesting in just the few years since the 1996-98 baseline period. It is likely that within 20 years most of the roadless areas identified on private commercial timberlands will be gone or greatly reduced in size. At that time, the only areas remaining will be those intentionally maintained on public or private conservation land or those where topography limits road construction and harvesting. The conservation value of the areas identified in this study varies widely. Conditions may range from notably mature forest that saw only limited selective harvesting many decades ago to midsuccessional forest of no unique ecological significance. While roadless condition is important when assessing conservation priorities, this information is most valuable when combined with other sources of information. These include on-the-ground assessment to determine current forest condition, the presence of features of high ecological or recreational value, and the position of the area relative to existing conservation lands or other high-value areas.

Introduction The 26 million acre (10.5 million hectare 1 ) Northern Forest of Maine, New Hampshire, Vermont and New York (Fig. 1) is the largest expanse of relatively undeveloped forest in the eastern United States (Harper et al. 1990). The region is characterized by large private forest ownerships, with nearly half of the land in ownerships over 5,000 acres managed primarily for timber production. Many of the thousands of smaller private landowners also have timber production as a primary or secondary goal. Only about 17% of the region is in public ownership, with about 70% of this lying within the Adirondack State Park in New York 2. Because of the extent of large private ownerships, considerable portions of the region have no permanent population or public roads. The region s 1 million citizens live in rural towns and small cities, primarily along the southern edge of the region. However, despite the common description of the region as wilderness, it is in fact a heavily utilized forest. The region has been undergoing timber harvesting for over a century, and essentially no primary forest remains (Davis 1996). The lack of mature Figure 1. The Northern Forest region. forest is one of the primary concerns of the region s ecologists (Gawler et al. 1996, NHDRED 1995). An extensive network of public roads and private logging roads blankets the region. The 10 million acres of unincorporated land in Maine contains an estimated 2,000 miles of public road and 20,000 miles of private road, with an estimated 500 miles of new road being added each year (MDOC 1997). Over the last 15 years, the future of the region has been the subject of considerable public policy debate. Significant changes in land ownership, increasing competitive pressures on the region s forest products industry, and increasing public recreational use have led to growing concern about the region s ability to maintain the ecological and socioeconomic values associated with this large forested region. Concerns include inappropriate development, overharvesting, the lack of significant natural areas and mature forest, and the lack of economic opportunities for the region s citizens. From 1990 to 1994, the Northern Forest Lands Council, a quasi-public advisory board, studied the region, conducted studies, and heard public testimony, eventually releasing a set of recommendations for public policy steps designed to conserve the social, economic and ecological values associated with the region s forests (NFLC 1994). Among the 1 1 hectare equals 2.47 acres. 2 These figures include the entire White Mountain National Forest. In the original delineation of the Northern Forest study area by USFS, the WMNF was split along county lines and included only 270,000 of the forest's 770,000 acres.

recommendations were calls for the states to establish ecological reserve systems and consider public land acquisition programs for areas of high ecological or recreational value. Since 1994, the pace of change in the region has accelerated. Additional large land sales have taken place, with over 20% of Maine changing hands since late 1998. There has been a shift in land ownership away from large paper companies to institutional and private investors, some of whom have expressed an interest in pursuing land conservation deals. In addition, there has been a significant increase in public support for land conservation, as shown by the passage of a $50 million land conservation bond in Maine and the establishment of a statewide land conservation investment program in New Hampshire. Of particular interest to the region s ecologists and conservation community is the opportunity to increase the amount of reserve or wilderness land in the region. Currently only about 12% of the region is managed as natural area 3, with well over 90% of this in three large public ownerships (the Adirondack State Park, the White Mountain National Forest, and Baxter State Park in Maine). The Adirondack State Park encompasses about three-quarters of all the natural area in the region. Outside of these areas, only 1 to 2% of the region is managed as natural area. States across the region are actively implementing ecological reserve programs, and natural areas have been established as a component of several major land conservation transactions. Identification of roadless areas is a key component of a landscape-level conservation strategy focused on establishment of wilderness areas or ecological reserves (Noss 1992, 1995). These are areas that have had relatively less impact from human activity than surrounding lands, and thus are likely to contain ecosystems with higher levels of ecosystem integrity. They are highly valued for providing core areas of unfragmented mature interior forest habitat. In the western United States, roadless areas represent areas that have remained free from significant human manipulation (road construction, commercial timber harvesting, etc.) since European settlement of North America. In the eastern United States, roadless areas are less pristine, as essentially all land has seen significant human impacts. However, they are areas that have seen limited activity for at least several decades, and thus they represent some of the least heavily impacted portions of an otherwise heavily impacted landscape. This study was undertaken to identify residual roadless areas greater than 2,000 hectares (4,940 acres) in size in northern Maine, New Hampshire and Vermont. Outside of the White Mountain National Forest (USDA 1978, 2000), no comprehensive effort to identify roadless areas across the region has been undertaken. This information can be of considerable value to conservation organizations, land trusts, government agencies and local citizens working with willing landowners to identify appropriate areas for conservation action, in particular areas suitable for ecological reserves. 3 This figure includes both legally designated wilderness and natural areas and other areas do not have permanent legal designation, or which preclude timber harvesting but which allow some motorized use (such as the Adirondack State Park s Wild Forest classification).

Methods Study Area The 9.4-million-hectare area considered during this analysis includes most of the undeveloped portion of the Northern Forest of Maine, New Hampshire and Vermont (Fig. 2). New York was not included. The study area extended south of the Northern Forest boundary to the limit of available satellite imagery, reaching as far south as the Ossipee Mountains in central New Hampshire. Data sources The primary data sources used during this study were recent satellite imagery, historical forest clearing data developed by AMC in an earlier study, and published road atlases (Delorme 1996, 1999, 2000). Figure 2. Roadless evaluation study area. Dashed line represents Northern Forest boundary. Recent satellite imagery consisted of a combination of Landsat Thematic Mapper (TM) (25 meter resolution) and Indian Remote Sensing (IRS) (20 meter resolution) images covering the period 1996-1998. A total of eight images were used to cover the study area (Fig. 3): Satellite Date Path Row IRS 24Jul97 292 35 IRS 24Jul97 292 36 IRS 11Jun97 293 35 IRS 11Jun97 293 35 IRS 22Aug98 294 36 IRS 02Jun96 295 37 TM 22Jul96 12 29 TM 30Aug96 13 29 Figure 3. Satellite images used in this study

Historical forest clearing data was developed using a time series of Landsat Multi-Spectral Scanner (MSS) images (50-meter resolution) covering the periods 1973-74, 1978-79, 1984-85 and 1990-91. A simple unsupervised classification procedure was used to classify images from each time period into forest and open categories. Forest clearing was determined by overlaying the images and identifying pixels that changed from forest to open across subsequent time periods. This analysis was conservative in that it attempted to identify only areas that were truly cleared, while not identifying areas that may have been heavily harvested but which retained some residual overstory. While some of the identified area may represent natural disturbance, the great majority is due to timber harvesting. Clearing for development has also taken place, though this is likely to be concentrated around existing developed areas and represent a small proportion of the total. Delorme road atlases were used as the primary accessory source of road data, as these are updated on a regular basis and are generally considered the most accurate delineation of current roads. Unfortunately, this data is not available in digital form. The primary potential source of digital road data (USGS Digital Line Graph data) was not used, as this information is much more out of date. While suitable for identifying public roads, it does not accurately reflect either the location or classification of the extensive private logging roads that cover much of the study area. Procedures and Criteria Potential roadless areas of at least 2,000 hectares were identified from on-screen viewing of recent satellite imagery in an ArcView environment. Boundaries were digitized on-screen along bounding roads and around the edges of developed or cleared areas. Delorme atlas maps were used as a comparison to assist with the identification and classification of roads. The following criteria guided the delineation of these areas: Roads were considered significant enough to affect delineation if they either 1) were clearly visible on the satellite images, or 2) were designated as improved roads in the Delorme atlases. These latter roads were always distinguishable on the imagery, though sometimes they were only marginally visible. Roads designated as unimproved roads or trails in the atlas were not considered if they could not be clearly distinguished on the imagery. Railroad and transmission line corridors were considered to be roads. Boundaries were drawn to exclude areas of obvious development or recent harvesting on the 1996-98 imagery, as well as areas shown as having been cleared in the historical data. Boundaries were not drawn to exclude dead-end roads intruding into designated areas. However, these roads were cherry-stemmed out of the areas during the final buffering process (see below). The roads were included when calculating statistics for interior road density. The maximum allowable density of dead-end interior roads was arbitrarily set at 0.2 km/km 2 (0.12 mi/mi 2 ). Where several dead-end roads were in close proximity, boundaries were drawn to exclude them as a group, rather than individually cherry-stemming them. These roads were thus not included in interior road density calculations. Generally roads closer than 1 kilometer were collectively excluded, though the decision was subjective based on the length and arrangement of the roads.

Lakes were included within roadless areas if the great majority of their shoreline was roadless and including them did not create narrow strips of included land between roads and lakeshores. Where roads lay in close proximity to a significant portion of a lake s shoreline, the boundary was drawn along the roadless shoreline but the lake itself was excluded. Rivers were not considered as boundaries. Bottlenecks (i.e., the distance between opposite edges of an area, or between the ends of interior dead-end roads and area boundaries) of less than 1 kilometer were not allowed. For final boundary delineation, bounding and interior roads, as well as cleared and developed areas, were buffered by 100 meters to account for edge effects and potential unseen development along the roads. Bounding lakeshores were not buffered. The 2,000 hectare minimum size was determined on the area remaining after buffering. Caveats The following limitations must be kept in mind when interpreting the results of this evaluation: The major limitation is that designated areas may contain maintained or well-used roads that are not readily visible on the satellite imagery, and which are not accurately mapped or classified in the Delorme atlas. While roads on the extensive large commercial ownerships of Maine and northern New Hampshire are for the most part readily visible, roads in other areas (especially parts of Vermont) are less visible due to the greater maturity of the forest and generally narrower roads. The Delorme atlases for Vermont and New Hampshire use a different road classification system than the Maine atlas, so that roads with similar characteristics may have been treated differently across the study area. In general, the Vermont and New Hampshire atlases make a less precise delineation between improved and unimproved roads than does the Maine atlas. Decisions on whether unimproved roads were clearly visible on the satellite imagery (and thus should be considered in delineation) involved some subjective judgement. When in doubt, the authors erred on the side of including these roads in delineation. Decisions on whether clusters of interior dead-end roads should be excluded as a group or included and cherry-stemmed were also subjective; when in doubt such roads were excluded. In some areas excessive haze or cloud cover on the recent satellite imagery made boundary delineation less precise. This was primarily a problem in portions of the Downeast Maine, Saddleback-Sugarloaf, and Upper St. John River areas. In these cases greater reliance was put on the Delorme atlas and historical clearcutting information. While areas were drawn to exclude areas of obvious heavy harvesting, they may contain areas of less intensive harvesting not evident on the recent imagery or areas of heavier harvesting not identified as cleared in the historical data. In some cases, the question of whether harvesting was noticeable enough to warrant exclusion (or whether the pattern seen on the imagery actually represented harvesting) involved subjective determination and considerable uncertainty. Because human-generated clearing can not be distinguished from natural disturbance on the satellite imagery, some cleared areas created through natural disturbance may have been inappropriately excluded from consideration.

Results The procedure used in this study proved to be a relatively straightforward method to delineate roadless areas across the region. Despite the subjective element, for the most part roadless area boundaries could be clearly delineated along obvious roads, development or heavily harvested areas (Fig. 4 4 ). Boundaries that were less clear in their definition represent a small proportion of the total results. A total of 137 roadless areas were identified (Fig. 5 4 ), ranging in size from 2,000 to over 78,000 hectares (4,940 to 193,000 acres). In total they encompass over 990,000 hectares (nearly 2.5 million acres), or approximately 11% of the total study area. The greatest concentration of roadless area was found along the spine of the northern Green Mountains in Vermont, in the White Mountain region of New Hampshire, and around Baxter State Park in Maine. Areas were also scattered throughout the more remote portions of western and northern Maine. Far fewer areas were identified in Vermont east of the Green Mountains, New Hampshire north of the White Mountains, and southern and eastern Maine. 78% of the areas were less than 8,000 hectares in size, with a median size of about 4,300 hectares (Fig. 6). The largest area (78,127 hectares), encompassing the core of Baxter State Park and adjacent lands, was over 60% larger than the next largest area (49,646 hectares). 70 60 50 No. of Areas 40 30 20 10 0 2-4 4-8 8-16 16-32 32-64 > 64 Size Class (thousands of hectares) Figure 6. Distribution of roadless areas by size. Interior roads Only 27 of the areas had no interior dead-end roads. The maximum density of interior roads in any area was 0.17 km/km 2 (0.11 mi/mi 2 ). However, 60% of the areas had interior road densities of less than 0.05 km/km 2, and only 9 of the 137 areas had interior road densities above 0.10 km/km 2. 4 At end of paper.

Shape Interior dead-end roads can affect the ecological integrity of roadless areas by increasing the perimeter of the area and thus creating more edge in relationship to interior habitat. However, many intruding roads were excluded from roadless area delineation, especially if there were heavily harvested areas associated with them. The need to draw boundaries to exclude roads and harvested areas, combined with the cherry-stemming of included interior roads, resulted in roadless area boundaries that were occasionally quite convoluted. One commonly used metric of shape is the perimeter to area ratio (PAR). For a given area, PAR is minimized by a circle. All other things being equal, this is the ideal shape for a roadless area, as it minimizes the proportion of edge to interior habitat. However, PAR is strongly related to area for equivalent shapes, PAR decreases with increasing area. To evaluate shape complexity independent of area, we used Relative Perimeter to Area Ratio (RPAR) the relationship between PAR of the roadless area and PAR of a circle of equivalent area. Two areas in the White Mountain region illustrate the extremes in roadless area shape (Fig. 7). The complexity of the Mahoosuc area (reflected in its higher RPAR) is due to its being located along a complex system of mountain ridges, while excluding adjacent roaded valleys with extensive harvesting. Figure 7. Extremes in roadless area shape: Great Gulf (compact) and Mahoosuc Range (convoluted). The effect of shape was also measured by calculating the proportion of interior habitat for each area (the ratio between the size of the area and its size before applying the 100-meter buffer around edges and interior roads). Due to its complexity, the Mahoosuc area has a lower proportion of interior habitat than the Great Gulf area, but the difference is minimized by the greater size of the Mahoosuc area. There is a clear relationship between shape complexity (RPAR) and proportion of interior, but only when measured across a fairly narrow range of area, and only for areas less than 8,000 hectares in size. Across a wider range of areas, or when only larger areas are considered, there is no correlation between RPAR and proportion of interior, as

the size of the area becomes the dominant factor in determining proportion of interior. At its extreme this effect is illustrated by the very large Baxter State Park roadless area despite having by far the greatest absolute length of dead-end interior road and a very high RPAR, it also has the highest proportion of interior habitat. Thus the effect of shape on interior habitat is most important for smaller areas; as the size of the area increases both its shape and the presence of dead-end interior roads becomes less relevant. Because of the relatively large minimum size used in this study, all areas have at least 86% interior, and 70% have at least 90% interior. While compact shapes are important, size is the most important factor in providing interior habitat (in both the absolute and proportional sense). Relationship to public lands Not surprisingly, roadless areas are disproportionately concentrated on public conservation lands. Public lands 5 make up 10% of the study area but encompass 44% of the total roadless area. About 49% of public land but only 6.5% of private land was mapped as roadless. Though there was no overall correlation between the size of a roadless area and the proportion of that area in public ownership, there is a relationship between large roadless areas and large blocks of public land. Of the total area in blocks above the median size 4,300 hectares), 50% was public land, while of the total area in blocks below the median size only 22% was public land. The 15 largest blocks (those above 15,000 hectares in size) were all located at least partially on a large block of public land (Fig. 5 6, Table 1). Of these 15 areas, 11 were more than 50% public land, and 8 were more than 75% public land. Table 1. Location and summary data of 15 largest roadless areas. Public land tract around which roadless area is located Roadless area size (ha) % of roadless area on public land Baxter State Park (Mt. Katahdin) 78,100 69 White Mountain NF (Pemigewasset) 49,600 99 White Mountain NF (Sandwich Range) 30,700 98 White Mountain NF (Wild River) 29,000 95 MBPL Mahoosuc Tract 22,600 26 White Mountain NF (Presidential-Dry River) 20,900 97 MBPL Deboullie Tract 20,200 28 Baxter State Park (north end) 19,800 77 MBPL Bigelow Preserve 19,300 62 MBPL Nahmakanta Tract (Debsconeag Lakes region) 18,800 28 US Navy Redington Tract (Saddleback-Sugarloaf region) 17,300 34 Mount Mansfield State Forest 17,000 62 White Mountain NF (Carr Mountain) 16,700 84 White Mountain NF (Kilkenny) 16,400 87 Camel s Hump State Park 15,900 56 5 Public land ownership was determined from digital conservation lands data obtained from the Maine Office of GIS, the University of New Hampshire Complex Systems Research Center (GRANIT database), and the Vermont Center for Geographic Information. Only lands owned in fee by public agencies was included. Private conservation lands and lands covered by conservation easement were not included. 6 At end of paper.

Relationship to elevation There is also a strong relationship between elevation 7 and the location of roadless areas (Fig. 8). 12% of roadless area acreage is above 2700 in elevation, though these high-elevation lands comprise less than 2% of the study area. Lands above 1500 make up 16% of the study area but encompass about half of the roadless areas. Less than 5% of land below 1700 was mapped as roadless, compared with over 70% of land above 2700 and essentially all land above 3700 8. This result is not surprising human settlement and development, as well as the location of logging roads and most harvesting activity, is concentrated in the gentler and more accessible terrain of lower elevation. 10 9 8 7 Figure 8. Relationship between roadless areas and elevation Percent of area (study area) Percent of area (roadless areas) Percent roadless 125 100 Percent of area 6 5 4 75 50 Percent roadless 3 2 25 1 0 0 1-100 300-400 600-700 900-1000 1200-1300 1500-1600 1800-1900 2100-2200 2400-2500 2700-2800 3000-3100 3300-3400 3600-3700 3900-4000 4200-4300 4500-4600 4800-4900 5100-5200 5400-5500 5700-5800 6000-6100 Elevation range (feet) Loss of roadless areas After the initial designation of areas was completed, a more recent satellite image 9 from the summer of 1999 was obtained covering northern New Hampshire, northeastern Vermont and northwestern Maine. The 1999 image overlapped with the two 1996 Landsat TM images (Fig. 9). 7 Elevation data was derived from U.S. Geological Survey 1:100,000 scale Digital Elevation Model data. 8 The exceptions are the Auto Road, Cog Railway and summit facilities on Mount Washington, the summit road and facilities on Mount Mansfield, and the upper reaches of several downhill ski areas. 9 Indian Remote Sensing image 291/37

This area encompassed all or parts of 41 roadless areas. Upon examining these areas on the 1999 imagery, 15 were found to have been impacted by roadbuilding or harvesting between 1996 and 1999. Approximately 1525 hectares were lost from these areas, or less than 1% of the total area. Individual lost areas ranged from 1 to 300 hectares. Not all of this lost area was actually impacted; in some cases a road or harvest area pinched off a larger section. While this rate of loss is quite low, it does indicate the erosion of these areas that is slowly but surely taking place. However, this assessment undoubtedly underestimates the overall rate of loss. Roadless acreage in the area of comparison area is dominated by large areas on the White Mountain National Figure 9. Area of overlapping 1996 and 1999 satellite imagery and roadless areas assessed for loss of area. Forest (where minimal road building is taking place), as well as blocks in relatively rugged topography that create long-term limitations on road-building. In theory it would be possible to comprehensively assess loss of roadless areas by conducting a region-wide roadless area mapping based on the earlier Landsat MSS imagery used in the assessment of historical forest clearing. However, this imagery has much lower spatial and spectral resolution than the more recent imagery; roads and harvest areas that are clearly visible on TM and IRS images are not visible on the MSS imagery. In addition, the historical forest clearing data would not reach as far back in time. Thus, it would not be possible to apply the same criteria or level of accuracy to an assessment based on MSS imagery. One area (the Debsconeag Lakes region southeast of Baxter State Park) does provide an opportunity to evaluate a specific case of roadless area loss. Prior to 1990, this area (Fig. 10 10 ) had long been managed as a remote recreational area by its landowners, and had remained relatively untouched since the last round of harvesting many decades ago. The extent of the roadless area could be accurately mapped on 1991 MSS imagery, with the boundaries well defined by major roads, lakeshores, and heavy harvesting that took place in the 1970s and 1980s. Extensive roadbuilding and harvesting took place in this area in the first part of the 1990s, with the result that between 1991 and 1997 the size of the roadless area shrank from 30,400 hectares to its current 18,800 hectares. Discussion The northern New England region is facing a period of both unprecedented change and unprecedented opportunities for large-scale conservation. The information developed in this 10 At end of paper.

study was developed to assist public agencies, land trusts, conservation organizations, and local citizens in the identification of ecologically significant portions of the landscape and the development of specific conservation proposals. The use of satellite imagery, though time consuming, is a practical way to delineate roadless areas across a large region based on the best available information on current condition. It is especially useful where existing digital road data (such as USGS DLG data) is insufficiently accurate to represent existing road networks, as is the case with the extensive private commercial forest lands of northern New England. The primary limitation of this methodology is the expense associated with obtaining the necessary imagery 11 and the time involved in visually examining imagery from across a region and manually delineating the boundaries of each area. Some roadless area assessments have been based only on the location of roads and permanent development features. However, one of the primary goals of roadless area conservation is to maintain areas that contain (or have the potential to contain) mature unfragmented forest. Thus it is appropriate to exclude areas that have undergone recent heavy harvesting. Such areas are relatively visible on the recent imagery used in this assessment. The inclusion of historical forest clearing information provides an added temporal dimension and increases the ability to exclude areas of young forest from roadless area delineation, even though these areas may have regrown to the point that they are no longer be visible on the most recent imagery. The data used in this assessment allowed the exclusion of any area that had been cleared at any time since the late 1960s (approximately 30 years). This period is admittedly an artifact of when the first MSS imagery became available. However, one scientific advisory panel has identified 40 years as an appropriate age to distinguish young from relatively mature forests in this region (MCSFM 1996), thus the exclusion of forest stands younger than 30 years of age is ecologically appropriate. Unlike roadless areas in the west, the areas identified in this study do not contain primary forest. All have been harvested (or even cleared) in the past, and many contain unimproved, abandoned, or otherwise low-impact roads and logging trails. Some of these roads are open for on-going motorized recreational or forest management use. Areas may have undergone less intensive recent harvesting not evident on the satellite imagery. Thus it is more appropriate to characterize these areas as very low road density rather than truly roadless. However, they clearly represent some of the least heavily impacted parts of an otherwise heavily roaded and harvested landscape. This study provides a snapshot in time of an ever-changing landscape. If this study had been done ten or twenty years ago the extent of roadless areas would have been greater. In contrast, some areas identified in this study would not have been mapped at various points in the past, when the evidence of heavy harvesting in those areas was still evident. A classic example is the Pemigewasset valley of the White Mountain National Forest, heavily logged during the first half of the 20 th century but now one of the premier wilderness areas in the east. This study 11 The distribution of Landsat Thematic Mapper imagery has recently been decommercialized and is now handled through the U.S. Geological Survey s Eros Data Center. As a result of this change, the cost of this imagery has been greatly reduced.

recognizes the potential for roadless area restoration as well as loss. However, the opportunity for roadless area restoration through benign neglect is declining rapidly. In the past, logging trails were temporary features abandoned after their initial use. However, since the mid-1900s, with the decline of river drives and logging railroads and the rise of roadbased transportation, logging roads have become essentially permanent features of the landscape designed for long-term use. While much of the modern logging road network is in place, construction and upgrading of roads continues. Many of the areas identified in this study in Maine have undergone road construction and harvesting in just the few years since the 1996-98 baseline period. It is likely that within 20 years most of the roadless areas identified on private commercial timberlands will be gone or greatly reduced in size. At that time, the only areas remaining will be those intentionally maintained on public or private conservation land or those where topography limits road construction and harvesting. Restoration of roadless areas will still be possible, but only through active closure and restoration of roads for conservation purposes. The conservation value of the areas identified in this study varies widely. Some may have undergone only limited selective harvesting many decades ago, and thus contain forest that is notably mature when compared to the broader landscape. Other may have been harvested more heavily and contain mid-successional forest of no particular ecological significance. While roadless condition is important when assessing conservation priorities, this information is most valuable when combined with other sources of information. These include on-the-ground assessment to determine current forest condition, the presence of features of high ecological or recreational value, and the position of the area relative to existing conservation lands or other high-value areas. Acknowledgements Support for this work was provided by grants to the Appalachian Mountain Club from the R. K. Mellon Foundation, the Doris Duke Charitable Foundation, the Moriah Fund, and the Surdna Foundation. GIS software (Arc/Info, ArcView, Image Analysis and Spatial Analyst) used during this project was provided by donations from ESRI s Environmental Conservation Program.

Literature Cited Davis, M.B. (ed.). 1996. Eastern Old-Growth Forests: Prospects for Rediscovery and Recovery. Island Press, Washington, D.C. DeLorme. 1996. Vermont Atlas and Gazetteer. Delorme, Yarmouth, ME. DeLorme. 1999. New Hampshire Atlas and Gazetteer. Delorme, Yarmouth, ME. DeLorme. 1996. Maine Atlas and Gazetteer. Delorme, Yarmouth, ME. Gawler, S.C., J.J. Albright, P.D. Vickery and F.C. Smith. 1996. Biological Diversity in Maine. Maine Department of Conservation, Natural Areas Program, Augusta, ME. Harper, S.C., L.L. Falk and E.W. Rankin. 1990. The Northern Forest Lands Study of New England and New York. USDA Forest Service, Rutland, VT. Maine Council on Sustainable Forest Management. 1996. Sustaining Maine s forests: criteria, goals and benchmarks for sustainable forest management. Maine Department of Conservation, Augusta, ME. Maine Department of Conservation. 1997. Comprehensive land use plan for areas within the jurisdiction of the Maine Land Use Regulation Commission. MDOC Land Use Regulation Commission, Augusta, ME. New Hampshire Department of Resources and Economic Development. 1995. New Hampshire Forest Resources Plan Assessment Report. NH DRED, Division of Forest and Lands, Concord, NH. Northern Forest Lands Council. 1994. Finding Common Ground: Conserving the Northern Forest. NFLC, Concord, NH. Noss, R.F. 1992. The Wildlands Project land conservation strategy. Pp. 10 25 in: Wild Earth special issue. The Cenozoic Society, Inc., Canton, NY. Noss, R.F. 1995. Maintaining ecological integrity in representative reserve networks. World Wildlife Fund Canada, Toronto, Ontario. U.S. Department of Agriculture. 1978. Northern Appalachian New England state supplement to USDA Forest Service Environmental Impact Statement: Roadless area review and evaluation II (RARE II). USDA Forest Service, Milwaukee, WI. U.S. Department of Agriculture. 2000. Forest Service Roadless Area Conservation Final Environmental Impact Statement. USDA Forest Service, Washington, D.C.

Figure 4. A 7,900 hectare (19,600 acre) roadless area just west of Caucomgomoc Lake in northwestern Maine. Base image is IRS image 292/35, band 4 (mid-infrared) from July 1997. In this image mature forest is dark gray; roads and recent heavy harvests are white to light gray. Lakes are shown in blue.

Figure 5. Roadless areas in northern New England.

Figure 10. The Debsconeag Lakes roadless area. The crosshatched area represents the roadless area as mapped in this study; the uncrosshatched area represents previously roadless area lost to roadbuilding and clearcutting between 1991 and 1997.