Home | Archives | About | Login | Submissions | Notify | Contact | Search
 ACE-ÉCO Home > Vol. 2, No. 2 > Art. 12

Copyright © 2007 by the author(s). Published here under license by The Resilience Alliance.
Go to the pdf version of this article

The following is the established format for referencing this article:
Harris, J. B. C., and D. G. Haskell. 2007. Land cover sampling biases associated with roadside bird surveys. Avian Conservation and Ecology - Écologie et conservation des oiseaux 2(2): 12. [online] URL: http://www.ace-eco.org/vol2/iss2/art12/


Research Papers

Land Cover Sampling Biases Associated with Roadside Bird Surveys
 
Biais d’échantillonnage des types de milieux associés aux dénombrements d’oiseaux nicheurs le long des routes

J. Berton C. Harris 1 and David G. Haskell 1


1University of the South



ABSTRACT


Roadside surveys such as the Breeding Bird Survey (BBS) are widely used to assess the relative abundance of bird populations. The accuracy of roadside surveys depends on the extent to which surveys from roads represent the entire region under study. We quantified roadside land cover sampling bias in Tennessee, USA, by comparing land cover proportions near roads to proportions of the surrounding region. Roadside surveys gave a biased estimate of patterns across the region because some land cover types were over- or underrepresented near roads. These biases changed over time, introducing varying levels of distortion into the data. We constructed simulated population trends for five bird species of management interest based on these measured roadside sampling biases and on field data on bird abundance. These simulations indicated that roadside surveys may give overly negative assessments of the population trends of early successional birds and of synanthropic birds, but not of late-successional birds. Because roadside surveys are the primary source of avian population trend information in North America, we conclude that these surveys should be corrected for roadside land cover sampling bias. In addition, current recommendations about the need to create more early successional habitat for birds may need reassessment in the light of the undersampling of this habitat by roads.

RÉSUMÉ

Les programmes de dénombrement d’oiseaux tels que le Recensement des oiseaux nicheurs sont fréquemment utilisés pour estimer l’abondance relative des populations d’oiseaux. La précision des dénombrements effectués le long des routes dépend de la réprésentativité des types de milieux échantillonnés par rapport à leur proportion relative dans la région. Nous avons quantifié le biais existant dans l’échantillonnage des milieux situés en bordure de routes au Tennessee, É-U, en comparant les proportions représentées par chacun de ces types de milieux près des routes vs dans l’ensemble de la région. L’échantillonnage en bordure de routes a résulté en un échantillon biaisé des patrons à l’échelle de la région : certains types de milieux étaient sous-représentés ou sur-représentés près des routes. Les biais changeaient dans le temps, ce qui introduit un degré variable de distorsion dans les données. À l’aide de ces données sur le biais d’échantillonnage le long des routes et de données sur l’abondance d’oiseaux, nous avons simulé les fluctuations de populations de cinq espèces d’un intérêt particulier pour l’aménagement. Ces simulations ont indiqué que les dénombrements le long des routes peuvent résulter en une impression faussement négative des tendances des populations des espèces de début de succession et des espèces synanthropiques, tandis que les résultats sont fiables pour les espèces de fin de succession. Puisque les dénombrements effectués le long des routes constituent la principale source d’information sur les tendances des populations d’oiseaux en Amérique du Nord, nous concluons que ces dénombrements devraient être corrigés pour les biais identifiés dans la représentativité des types de milieux échantillonnés. De plus, les recommandations actuelles au sujet de la création d’habitat de début de succession devront être réévaluées en fonction du sous-échantillonnage de ce type d’habitat le long des routes.


Key words: Breeding Bird Survey; early successional; industrial forestry; point count; population estimate.



INTRODUCTION


Effective conservation or management of any species requires that reliable information be available about the population trends of the species in question (Bart et al. 2004a, Sauer et al. 2005b). This information is usually obtained by sampling a population over time and then estimating overall population trends from these samples. Thus, the reliability of any conclusions about population trends depends critically on the quality of the sampling methods used to estimate these trends (Link and Sauer 1998). Sampling methods that introduce biases could result in over- or underestimates of population size. These potential biases are particularly problematic if the extent of the bias changes over time, introducing variable levels of distortion into the data. Temporal variation in bias could produce spurious trends or mask true trends and would make correcting for the bias more challenging. Bird surveys have used a variety of sampling schemes to quantify populations. These range from counting nests in seabird colonies, to counting migrating raptors, to conducting playback surveys for owls (Bird Studies Canada 2004). Each method has its own associated set of biases. Bart (2005) reviews these survey methods and concludes that the correction of bias is essential for the reliable use of survey results.

The Breeding Bird Survey (BBS) is a survey of North American bird populations that has been conducted along approximately 3000 40 km long routes since 1966 (Sauer et al. 2005a, K. Pardieck personal communication; the exact number of routes surveyed varies from year to year). BBS results are widely used to make conservation plans and to assess the status of the North American avifauna. Interpretations of the BBS form the basis of federal, state, and local policies and influence conservation funding decisions (Robbins et al. 1989, Sauer and Droege 1990, Peterjohn and Sauer 1994, Brown et al. 1995, Villard et al. 1995, Curnutt et al. 1996, Flather and Sauer 1996, Sauer et al. 1996, Koenig 1998, Brawn et al. 2001, Rich et al. 2004). Recently, for example, BBS data have been used to suggest that early-successional bird species in North America are experiencing problematic population declines, and recommendations have been made to intensively manage and create new early-successional habitat to benefit these species (Brawn et al. 2001, Hunter et al. 2001, DeGraaf and Yamasaki 2003, Dettmers 2003, Litvaitis 2003, Oehler 2003).

The BBS is subject to a variety of biases that must be accounted for to ensure reliable results (O’Connor et al. 2000, Bart et al. 2004a, Sauer et al. 2005b). Recent studies evaluating the use of BBS data underscore the need for quantitative assessment of these biases (O’Connor et al. 2000, Bart et al. 2004a, Sauer et al. 2005b, Thogmartin et al. 2006). Past studies have documented biases that affect the BBS in four categories: differences in observer detection rates, bias caused by analytical methods, differences between on- and off-road point counts, and differences in land cover sampling (Table 1).

Three studies have analyzed how well the land cover proportions sampled by roadside surveys represent the land cover proportions of the region under study (Bart et al. 1995, Keller and Scallan 1999, Betts et al. 2007). This bias is critical to the reliability of the BBS and other roadside bird surveys such as the Northern Region Landbird Monitoring Program and the American Woodcock singing-ground survey (Bird Studies Canada 2004). If roads are disproportionately more common in some land cover types than in others, then roadside surveys may provide a biased estimate of overall population levels. If the distribution of land cover types in relation to roads changes over time, then these biases could produce spurious or variably distorted population trends.

Both Bart et al. (1995) and Keller and Scallan (1999) compared land cover near roads to land cover away from roads. At 27 locations in western Ohio that were not actual BBS routes, Bart et al. (1995) used aerial photographs to compare forest cover at three levels from roads: 0–140 m from roads, 140–280 m from roads, and in the surrounding 21 km² region; they found very low bias in estimates of land cover change (< 1%) along roads compared to away from roads, although there was significantly less forest cover near roads than in the surrounding region. Keller and Scallan (1999) used aerial photographs to compare land cover types within 200 m of BBS routes to land cover on a strip of land located 200–1600 m away from BBS routes in Maryland (n = 28 routes) and Ohio (n = 25 routes). They found that land cover changes near routes generally also occurred away from routes, although significantly more urban cover occurred along BBS routes than in the surrounding landscape along Maryland routes, but not along Ohio routes (Keller and Scallan 1999). These studies provide a preliminary assessment of potential roadside sampling land cover biases. However, because they do not compare roadsides to the entire surrounding region, it is not clear to what extent the results are dependent on the subsample of nonroad land cover that they analyze. In addition, these studies do not quantify how bird population trends might be affected by roadside sampling. Because bird species differ in their land cover preferences, the sampling of species that live in land cover types that are over- or undersampled by roads will be differentially affected by roadside land cover sampling bias.

Betts et al. (2007) studied the question of roadside sampling bias by relating land cover change to population trends of one forest-dwelling bird species, the Blackburnian Warbler (Dendroica fucsa), in New Brunswick. They used aerial photographs to compare the extent of mature forest within 150 m of BBS routes to forest cover in surrounding 1° blocks of latitude and longitude from 1974 to 2001. Betts et al. (2007) found that forest loss was more rapid in the blocks than along BBS routes from 1974 to 1985 and that forest loss was equally rapid in the comparisons from 1985 to 2001. They also found that the Blackburnian Warbler showed a decline corresponding to the loss of mature forest from 1985 to 2001, but not from 1974 to 1985 (Betts et al. 2007). These results suggest that the decline in the bird species from 1974 to 1985 was underestimated by the BBS because of the survey’s roadside land cover sampling bias. Betts et al. (2007) demonstrated how roadside bias can affect estimates of the population trend of one mature-forest species. They did not, however, examine how these biases might relate to other species or land cover types.

We analyzed land cover on the southern Cumberland Plateau, Tennessee, and in the state of Tennessee, USA, to quantify roadside bias in the sampling of land cover. We tested the following predictions: (1) roadside surveys on the Cumberland Plateau and BBS routes in Tennessee give a biased estimate of land cover across the entire landscape, (2) the extent of these biases changes from year to year based on temporal changes in the distribution of land cover over the landscape, and (3) these biases influence the magnitude or direction of simulated bird population trends. Our study differs from previous work in that: we quantified how representative roadside samples compare to the entire region, we analyzed these patterns in all of the prominent land cover types of the region, and we analyzed the effects of the bias on simulated bird population estimates and trends for five bird species of conservation importance.



METHODS


We analyzed land cover along roads and away from roads on the southern Cumberland Plateau in eastern Tennessee, USA, using both fine- and coarse-resolution land cover databases. The Cumberland Plateau is a sandstone plateau that extends from Alabama to Kentucky and averages 600 m in elevation. Oak (Quercus spp.)-hickory (Carya spp.) forests dominate the region, with interspersed areas of pine plantations and townships; McGrath et al. (2004) and Haskell et al. (2006) provide detailed descriptions of the sampling area on the plateau. The fine-resolution land cover database contained vector (i.e., geographical information stored as digital polygons) land cover information for a seven-county area of the surface of the southern Cumberland Plateau that was derived from aerial photographs (1:24,000 scale) for 1980, 1990, 1997, 2000 (Evans et al. 2002, McGrath et al. 2004), and 2003 (J. Evans unpublished data). Every block of land cover that fell into the six categories listed below was digitized by hand from aerial imagery, regardless of the size of the land cover block. The overall accuracy based on ground-truthing of the vector layer was 80.6%, whereas the accuracy was > 90% for all pine plantations and 85% for mature native forest (Evans et al. 2002). This layer divided land cover into the following six categories (Table 2): mature native forest, thinned native forest, early-stage pine plantation, middle-stage pine plantation, mature pine plantation, and residential/rural. The coarse-resolution raster, i.e., geographical information stored as pixels, database was derived from LandSat Thematic Mapper satellite imagery (30-m pixel size; Tennessee Spatial Data Server 2005); the average accuracy of the raster layer at the per-patch level was 81% and at the per-pixel level was 62% (United States Geological Survey 2007). The raster layer divided land cover into 11 categories (Table 2): upland deciduous forest, upland mixed forest, upland coniferous forest, urban/developed, pasture/grassland, row crop, nonvegetated (e.g., barren land, strip mine, rock quarry), open water, forested wetland, nonforested wetland, and undefined (satellite image obstructed by clouds). The undefined class represented 0.007% of the landscape and was omitted from the analysis.

We created buffers of 50, 200, and 400 m around roads on the southern Cumberland Plateau for all named public roads (United States Census Bureau 2001). These distances represent the range of distances from which birds can be heard from Breeding Bird Survey (BBS) points and span most of the range previously used to estimate bird populations from BBS data (Rosenberg and Blancher 2005). In addition to analyzing land cover near roads, we used the coarse-resolution database to compare land cover near the actual BBS routes in Tennessee (n = 50) to the land cover proportions of the entire state (United States Geological Survey 2002). We studied how well the BBS samples land cover types at the state level for two reasons: conservation planning is often undertaken at the state level (e.g., Association of Fish and Wildlife Agencies 2007); and the BBS reports its findings by state, as well as by U.S. Fish and Wildlife Service multistate regions and physiogeographic provinces. Three of the BBS routes that we analyzed (i.e., Soddy 82030, McFarland 82035, and Walland 82039) are now no longer surveyed. Digitizing errors in the BBS database added artificial route length to two routes (i.e., Cades Cove 82904 and Newfound Gap 82903). We deleted these artificial segments before making the buffers. The BBS routes were all along roads in the named public roads database described above.

For all analyses, we measured the bias in land cover proportions near roads by calculating percentage of over- or underrepresentation of land cover near roads.

Equation 1(1)

where Biasx,y,z is the percent bias in land cover x for buffer width y in year z; Px,y,z is the proportion of land cover x in buffer width y in year z; and PT,x,z is the proportion of land cover x in year z over the entire region. Positive values indicate overrepresentation of a land cover in buffers along roads; negative values indicate underrepresentation.

Roadside surveys have been used to estimate the absolute population sizes of birds (Rich et al. 2004) and trends in bird populations over time (e.g., Robbins et al. 1989). Both of these estimates may be biased if surveys from roads do not representatively sample the entire landscape. Exactly how the estimates are affected depends on: the magnitude of the mis-sampling of land cover by roadside buffers, whether and how this misrepresentation changes over time, and how land cover proportions on the whole landscape change over time relative to those in roadside buffers. To explore how these factors might affect estimates of bird population sizes and trends, we modeled estimates of bird populations using five bird species. Four of the five species were the four top-ranked “species of continental importance” in the North American Landbird Conservation Plan (Rich et al. 2004); their field abundance in our study region was at least 0.1 birds detected per hectare in Haskell et al.’s (2006) analysis. These four species were Prairie Warbler (Dendroica discolor), Kentucky Warbler (Oporornis formosus), Eastern Towhee (Pipilo erythrophthalmus), and Wood Thrush (Hylocichla mustelina). These species are not only of high conservation interest, but they span a range of land cover preferences and therefore illustrate how roadside biases may affect surveys of species with different ecological requirements. The fifth species was European Starling (Sturnus vulgaris), a synanthropic (positively associated with human settlements) exotic species that has negative effects on some native species and is thus of management interest (Cabe 1993). The simulated relative densities in each land cover type for these five species (Table 3) correspond to field estimates in this study area made from point counts (n = 503) conducted during the breeding season (Haskell et al. 2006) and analyzed using both a fixed-radius (50 m radius) estimate and a distance-sampling estimate. All counts were made by the same observer. The estimates used in this simulation were based on fixed-radius counts. The model was designed to provide general information about how roadside bias may affect estimates of bird populations. The model did not aim to provide information about actual historical population trends of birds in the region. Rather, we used simulated populations of these species as a modeling exercise and deliberately held constant all other factors that might affect bird populations to examine only the effects of changing roadside bias and land cover.

In the simulation, the modeled population estimate of each species in each year for the whole landscape was the population density of the simulated species (Table 3) in each land cover multiplied by the number of hectares of each land cover for that year.

Equation 2(2)

where Ss,z is the simulated number of individuals in the entire region for species s in year z; Ds,x is the density of species s in land cover x; and Ax,z is the area of land cover x in year z. The modeled population estimate of each bird species in each year for the simulations using buffers of 50, 200, and 400 m width is the estimated mean population density of the species in each land cover type multiplied by the number of hectares of each land cover class indicated by surveys limited to within the buffer widths.

Equation 3(3)

where Ss,y,z is the simulated number of individuals in the study region for species s in year z, estimated for buffer of width y; Ds,x is the density of species s in land cover x; T is the total area of the study region; Ax,y,z is the area of land cover x within buffer of width y in year z; and By,z is the total area within buffer of width y in year z. Therefore, the simulation allows the comparison of simulated populations across the whole landscape with the populations that would be revealed by roadside surveys. This provides a demonstration of the potential effects of roadside land cover sampling bias on bird population estimates.

We calculated simulated population trends using linear regression of population size against time, after checking normal probability plots. Because we analyzed data points from one region, not from multiple routes, our regression approach was simpler than linear route regression based on estimating equations (Link and Sauer 1994) and the locally weighted least squares method that uses nonlinear regression (James et al. 1996).



RESULTS


For all land cover types, roadsides provided a biased estimate of the actual land cover of the region. In general, this bias was higher for narrower buffers (Tables 4 and 5). In the fine-resolution land cover analysis for the southern Cumberland Plateau, residential/rural areas were oversampled by roads by up to 266%, mature native forest was undersampled by up to 25%, and pine plantations were undersampled by up to 80% (Table 4). In the coarse-resolution analysis, urban/developed, row crop, and pasture/grassland were overrepresented by roads by up to 866%, 55%, and 168%, respectively (Table 5). These land cover types are equivalent to the residential/rural land cover category in the fine-resolution data set. Upland deciduous forest and upland coniferous forest were underrepresented by up to 41% and 44%, respectively; on the southern Cumberland Plateau upland coniferous forest is mostly pine plantation, whereas upland deciduous forest is mature native oak-hickory forest (Table 2).

The extent of these biases changed over time because of large-scale changes in the distribution of land cover. The percentage misrepresentation of residential/rural areas changed by as much as 128% over time. Mature native forest bias changed by as much as 8% and pine plantation bias changed by as much as 63% over time (Table 4).

In general, the same land cover types that were overrepresented in the analyses of the Cumberland Plateau were overrepresented around Tennessee Breeding Bird Survey (BBS) routes and land cover types that were underrepresented in the plateau analyses were underrepresented around BBS routes (Table 5). Urban/developed areas around BBS routes were overrepresented by up to 69%, and row crop and pasture/grassland were overrepresented by up to 23% and up to 34%, respectively. Upland deciduous forest was underrepresented by up to 32%. Upland coniferous forest was either overrepresented or underrepresented, depending on the buffer width. This upland coniferous forest consists of pine plantations in some regions of Tennessee (e.g., on the Cumberland Plateau) and natural coniferous forest in others (e.g., high elevations in the Smoky Mountains).

The extent to which these biases affected estimates of population sizes and population trends differed by species. When we simulated populations of birds on the southern Cumberland Plateau by combining our measurements of roadside bias with simulated data on bird densities in different land cover types within the study area, we found that both estimates of population size and estimates of population trends differed if the estimates were made for roadside buffers compared to the surrounding region.

Roadside surveys overestimated the population size of European Starling by several times (range of overestimate: 60.2–265.7%; Table 6). The overestimate was most severe for 50-m buffers, but even for 400-m buffers, the overestimate ranged from 60.2 to 96.3% depending on the year. The magnitude of the bias decreased over time for all three buffer widths as the extent of overestimation of residential areas declined (Table 4). The population sizes of Eastern Towhee and Wood Thrush were also generally overestimated, but to a lesser degree than for European Starling: up to 65.7% for Eastern Towhee and up to 6.5% for Wood Thrush. The magnitude of the overestimate for Eastern Towhee declined through time and turned into a slight underestimate for the 400-m buffer in 2003, whereas the magnitude of the overestimate increased through time for Wood Thrush. The population sizes of Kentucky Warbler and Prairie Warbler were both underestimated by roadside surveys (Table 6). The magnitude of this underestimate increased through time for Prairie Warbler, but decreased for Kentucky Warbler. The underestimate was more severe for Prairie Warbler (range: –11.8 to –41.8%) than for Kentucky Warbler (range: –2.1 to –15.8%).

For simulated estimates of population trends (i.e., changes in population size through time), roadside surveys gave biased results for all species (Table 7). These differences were statistically significant in four of five species. The population increases of Eastern Towhee, European Starling, and Prairie Warbler were underestimated by roadside surveys by 1.20–1.87, 1.03–2.02, and 1.69–3.17 times, respectively. The population decline of Wood Thrush was underestimated (i.e., estimated to be less negative) by roadside surveys by 1.43–1.65 times, whereas the decline of Kentucky Warbler was very slightly exaggerated (i.e., estimated to be more negative) by roadside surveys.



DISCUSSION


Roadside land cover sampling biases

Our data indicate that roadside surveys in the study area sampled land cover types in different proportions than they occurred in the surrounding region (Tables 4 and 5). This pattern was not limited to roads on the southern Cumberland Plateau; land cover types were misrepresented in similar ways by Breeding Bird Survey (BBS) routes across Tennessee. The magnitude of these biases was substantial. The BBS in Tennessee, for example, undersampled upland deciduous forest, the largest land cover in the state, by nearly one-third and oversampled urban/developed areas by two-thirds based on 50-m buffers around roads.

The biases that we measured were qualitatively similar in the fine- and coarse-grained analyses that we conducted in the same region. This suggests that coarse-grained data such as the publicly available interpretations of LandSat Thematic Mapper satellite data (United States Geological Survey 2006) may be sufficient to detect, quantify, and possibly correct biases in roadside sampling. We caution, however, that analyses are needed from other regions before this conclusion can be generalized. LandSat data are generally broken into a few broad thematic categories that, in some regions, may obscure important land cover differences. For example, on the southern Cumberland Plateau, the upland deciduous forest land cover category is composed of native vegetation and the upland coniferous forest is mostly composed of exotic pine plantations (Table 2). However, in other areas, these categories may combine native forest and exotic plantations, land cover types that are likely to have very different bird communities. In addition, publicly available raster land cover data may contain significant sources of error (Thogmartin et al. 2004, United States Geological Survey 2007). The quantification of roadside bias will be most effective when biases for different land cover types can be teased apart using accurate fine-resolution land cover data.

Because the fine-resolution land cover data for the southern Cumberland Plateau extended over several years, we were able to estimate the extent of temporal changes in roadside sampling bias. Such temporal analyses were not possible for the coarse-resolution databases that covered just 1 yr. On the southern Cumberland Plateau, we found that the extent of roadside biases changed over time because of large-scale changes in the distribution of land cover (Table 4). Two large-scale processes appear to drive these trends. First, extensive areas of native forest were cleared for pine plantations during the time period covered by our study (Evans et al. 2002, McGrath et al. 2004). These conversions of land cover occurred, on average, away from roads, but the exact distance from roads of new forest clearings varied over time, creating varying levels of undersampling by roads. Second, low-density housing development sprawled out from existing towns, converting native forest into residential areas (Evans et al. 2002, McGrath et al. 2004) and making native forest proportionally less common near roads at the end of the study period than at the start. In addition, the decrease in the percentage of oversampling of residential areas by roads suggests that the new residential areas were less intensely permeated by roads than were older townships.

Effects on estimates of bird populations

The magnitude of the biases that we report seem large, but their effects on surveys of bird populations cannot be immediately inferred from the examination of data on land cover bias (Tables 4 and 5). To examine how these biases might affect estimates of bird population sizes and population trends, we combined our measurements of roadside bias with simulated data on bird densities in different land cover types within the study area. These simulations held all other variables constant to examine the effects of roadside bias only. We found that roadside bias influenced both estimates of population size and estimates of population trends, but that the extent of this influence differed with the land cover preferences of the species under simulation. Before discussing the simulation results, we emphasize that this simulation provides modeled information about the potential effects of changing roadside bias and land cover and does not provide information about actual historical population trends of birds in the study region.

For estimates of population size (Table 6), we found that simulated European Starling was overestimated by 60.2–265.7%, depending on the year and buffer size used along roads. Thus, the biases that we found in roadside surveys (Tables 4 and 5) can potentially have substantial effects on the conclusions of bird surveys. The population sizes of Eastern Towhee and Wood Thrush were also overestimated by roadside surveys, although by a lesser degree than for European Starling. For European Starling and Eastern Towhee, the magnitude of the overestimate decreased through time, but for Wood Thrush the magnitude increased. The differences in these trends through time are caused by the different land cover preferences of the species: the overestimate of European Starling and Eastern Towhee is strongly influenced by declining oversampling of residential/rural areas, whereas Wood Thrush population estimates are affected by mature native forest, thinned native forest, and residential/rural areas, which are land cover types that showed conflicting trends of oversampling through time (Table 4). In contrast, the population sizes of Prairie Warbler and Kentucky Warbler were underestimated by roadside surveys (Table 6). The magnitude of the underestimate for Kentucky Warbler fluctuated through time, but the estimate for Prairie Warbler increased in magnitude because of the increasing underestimation of this species’ preferred land cover types: thinned native forest and middle-stage pine plantations.

Not only are overall population estimates biased, but the changing magnitude of these biases distorts estimates of population trends. Our simulations found that these distortions vary considerably according to the land cover preferences of the species in question (Table 7). We found that roadside surveys significantly underestimated the population increase of European Starling by up to 2.57% per year depending on the buffer width. This misrepresentation was caused by the high but declining overpresentation of residential land cover types along roadsides in the study area.

Because of the shrinking area of mature native forest in the study region, the two late-successional species, i.e., Wood Thrush and Kentucky Warbler, both had simulated population declines. For Wood Thrush, these declines were significantly underestimated by nearly one-half by simulated roadside surveys. Roadside surveys slightly but significantly exaggerated the decline for Kentucky Warbler by 0.05% per year (Table 7). The differences between the degree of bias for these two species was caused by the different habitat preferences for immature forest habitat (e.g., residential areas and pine plantations) used in the simulation (Table 3).

Because large areas of native forest were converted to pine plantations during the study period, the simulated populations of both Prairie Warbler and Eastern Towhee increased by > 3% per year across the 23 yr of the study period. However, roadside surveys indicated a much smaller yearly population increase, especially for narrower buffers. The magnitude of the underestimate differed between the two species because they are found in different densities in residential areas. Because of the large standard error of the regression coefficients in the simulation (Table 7), these differences were statistically significant for Eastern Towhee in narrow buffers, but not for other buffer sizes or for any buffers sizes for Prairie Warbler.

We draw two main conclusions from the simulated estimates of population size and population trends. First, the roadside biases that we report (Tables 4 and 5) have the potential to produce substantial distortions in estimates of both population sizes (Table 6) and population trends (Table 7). Population sizes were particularly poorly estimated by roadside surveys in the cases of European Starling, Prairie Warbler, and Eastern Towhee. These species live in land cover categories that are either undersampled (early-successional forests) or oversampled (residential areas) by roadside surveys, causing a large difference between population size estimates based on roadside surveys and population size estimates based on the whole landscape. Population trends were likewise poorly estimated by roadside surveys for species that live in early-successional forests and residential areas. These trend estimates were affected by both the yearly magnitude of land cover bias and the extent to which this bias changed from year to year. The size of the roadside bias in these estimates of population trends was large enough in some cases to potentially produce serious distortions in trend estimates. For example, for both Prairie Warbler and European Starling, the annual trend estimate (i.e., percent population change per year) was > 2% per year smaller in roadside surveys than for the whole landscape (Table 7). Given that for actual BBS data the mean annual population trend of declining species in the BBS is –1.09% (± 0.12 SE, range: –4.9 to –0.1%) and of increasing species is 2.16% (± 0.28 SE, range: 0.1–12.4%), the biases that we found in our simulations are well within the range that could change the conclusions of analyses of BBS data. It is important to note that these BBS means were obtained from the survey-wide 1966–2005 trends published on the BBS website using only those species with the highest reliability index, i.e., species with at least 14 samples over the long term, with moderate precision, and with moderate abundance on routes (Sauer et al. 2005a). We provide this information as a context for our results only and not as a quantitative analysis of actual BBS trends. Because temporal data on roadside bias were only available for a subregion of the Cumberland Plateau, and not the whole of Tennessee, our simulation focused on this subregion. Only part of one BBS route passes through this subregion; therefore, we cannot compare our simulated trends to trends reported by the BBS. Such comparisons will be possible when more detailed data on changing land cover at a large scale become available.

Our second main conclusion from the simulation is that our data suggest that the effect of roadside land cover sampling bias differs markedly among species. Each of the five species that we analyzed had its own set of distortions of population sizes or population trends. This heterogeneity results from the different land cover preferences of the species in question. Thus, although Wood Thrush and Kentucky Warbler both nest primarily in mature forest, their differing preferences for pine plantations and residential areas (Table 3) caused our simulations of their populations to respond quite differently to roadside bias. Similarly, although Eastern Towhee inhabits early-successional habitat and some residential areas, the roadside bias that we quantified in the simulation differed between Eastern Towhee and Prairie Warbler (which is rarely found in residential areas) and European Starling, which is only found in residential areas. Thus, correcting for roadside bias in BBS surveys will require data on the abundance of each species in each land cover type.

Our simulation assumed a linear relationship between land cover area and bird population size. In reality, bird populations often show nonlinear responses to changes in land cover (e.g., Meents et al. 1983). These nonlinear responses will affect roadside bias in complex ways, depending on the exact nature of the nonlinearity and the extent of habitat fragmentation. For example, if bird populations become disproportionately low in small habitat fragments compared to large fragments, then the extent of the roadside bias will depend on whether roads give a representative sample of habitat fragment sizes on the landscape. This question is beyond the scope of our study, but previous work suggests that roads may not provide random samples of fragment sizes, partly because roads are themselves causes of fragmentation (Heilman et al. 2002).

General implications

Our quantification of bias has two main implications for conservation biologists and land managers. First, our study, along with that of Betts et al. (2007), suggests that roadside surveys give biased samples of the landscape in some regions of North America. In contrast, Bart et al. (1995) and Keller and Scallan (1999) tentatively concluded that roadside survey land cover sampling bias is relatively minor. We recommend that future studies further quantify these biases in other regions and evaluate ways of correcting biases in existing survey data. The correction of this bias is conceptually straightforward: species abundance in each year can be weighed by the extent to which each species’ preferred land cover types are sampled by roads. However, the empirical data that are required to make these corrections are not yet available: temporal and spatial variation in the extent of roadside bias has not been quantified for most regions and data on bird abundance across land cover types are scattered throughout the literature or are unavailable. Nevertheless, these biases must be removed if the scientific community is to move toward statistically rigorous methods of interpreting BBS data.

Second, the biases that we quantified in the study area suggest that conservation recommendations based on uncorrected roadside survey data may need reinterpretation. For example, early-successional bird species have been the focus of a number of studies that recommend increasing rates of forest disturbance to create habitat for these species (Brawn et al. 2001, Hunter et al. 2001, DeGraaf and Yamasaki 2003, Dettmers 2003, Litvaitis 2003, Oehler 2003). However, we found that in the study area, industrial forestry activities that created large areas of early-successional habitat were poorly sampled by roads, suggesting that populations of species that inhabit these areas are faring much better than roadside surveys suggest. Unquantified reports of similar biases in other regions (Hagan et al. 1997) indicate that the problem of undersampling early-successional habitat on industrial timberland may be widespread within the BBS database. Betts et al. (2007) also emphasize that industrial timberlands tend to be located away from roads. However, biases may be weaker in areas that have little industrial forestry (Bart et al. 1995, Keller and Scallan 1999). If these results are confirmed elsewhere, the high priority currently assigned to early-successional bird species may need to be reassessed.

Our analysis focused on the effects of roadside bias on surveys of breeding birds. However, because roads provide a convenient means of access to many habitats, roadside surveys have been used to sample taxa other than birds. For example, the spread of invasive dogwood anthracnose in the southern Appalachians has been monitored through the use of roadside surveys (Sherald et al. 1996). Roadside surveys have also been used to quantify populations of amphibians (Royle and Link 2005) and mammals (Drake et al. 2005). The biases that we report here suggest that the results of these roadside surveys should be interpreted in light of the nonrandom distribution of roads on the landscape.



RESPONSES TO THIS ARTICLE


Responses to this article are invited. If accepted for publication, your response will be hyperlinked to the article. To submit a response, follow this link. To read responses already accepted, follow this link




ACKNOWLEDGMENTS

Support for JBCH was provided by the University of the South through a Harry C. Yeatman scholarship. We thank the Biology Department and the Landscape Analysis Laboratory at the University of the South for providing logistical support. We thank N. Hollingshead, K. Willis, and J. Evans for technical advice, and A. Chenault, M. Harris, D. McGrath, V. Moye, L. Reid, S. Vance, M.-A. Villard, K. Zigler, and four anonymous reviewers for assistance with the manuscript.



LITERATURE CITED


Association of Fish and Wildlife Agencies (AFWA). 2007. State wildlife action plans. Available online at: http://www.wildlifeactionplans.org.

Bart, J. 2005. Monitoring the abundance of bird populations. Auk 122:15-25.

Bart, J., J. B. Buchanan, and B. Altman. 2005. Improving the breeding bird survey. Pages 771-776 in C. J. Ralph and T. D. Rich, editors. Bird conservation implementation and integration in the Americas: Proceedings of the Third International Partners in Flight Conference. U.S. Department of Agriculture Forest Service, General Technical Report PSW-GTR-191.

Bart, J., K. P. Burnham, E. H. Dunn, C. M. Francis, and C. J. Ralph. 2004a. Goals and strategies for estimating trends in landbird abundance. Journal of Wildlife Management 68:611-626.

Bart, J., B. Collins, and R. I. G. Morrison. 2003. Estimating population trends with a linear model. Condor 105:367-372.

Bart, J., B. Collins, and R. I. G. Morrison. 2004b. Estimating trends with a linear model: reply to Sauer et al. Condor 106:440-443.

Bart, J., M. Hofschen, and B. G. Peterjohn. 1995. Reliability of the Breeding Bird Survey: effects of restricting surveys to roads. Auk 112:758-761.

Bart, J., and J. D. Shoultz. 1984. Reliability of singing bird surveys: changes in observer efficiency with avian density. Auk 101:307-318.

Betts, M. G., D. Mitchell, A. W. Diamond, and J. Bêty. 2007. Uneven rates of landscape change as a source of bias in roadside wildlife surveys. Journal of Wildlife Management 71:2266-2273.

Bird Studies Canada. 2004. North American bird monitoring projects database. Commission for Environmental Cooperation and Bird Studies Canada, Port Rowan, Canada. Available online at: http://www.bsc-eoc.org/nabm/index.jsp.

Brawn, J. D., S. K. Robinson, and F. R. Thompson. 2001. The role of disturbance in the ecology and conservation of birds. Annual Review of Ecology and Systematics 32:251-276.

Brown, J. H., D. W. Mehlman, and G. C. Stevens. 1995. Spatial variation in abundance. Ecology 76:2028-2043.

Cabe, P. R. 1993. European Starling (Sturnus vulgaris). In A. Poole and F. Gill, editors. Birds of North America, number 48. Academy of Natural Sciences, Philadelphia, Pennsylvania, USA, and American Ornithologists’ Union, Washington, D.C., USA. [online] URL: http://bna.birds.cornell.edu/BNA/.

Curnutt, J. L., S. L. Pimm, and B. A. Maurer. 1996. Population variability of sparrows in space and time. Oikos 76:131-144.

DeGraaf, R. M., and M. Yamasaki. 2003. Options for managing early-successional forest and shrubland bird habitats in the northeastern United States. Forest Ecology and Management 185:179-191.

Dettmers, R. 2003. Status and conservation of shrubland birds in the northeastern US. Forest Ecology and Management 185:81-93.

Drake, D., C. Aquila, and G. Huntington. 2005. Counting a suburban deer population using forward-looking infrared radar and road counts. Wildlife Society Bulletin 33:656-661.

Emlen, J. T., and M. J. DeJong. 1992. Counting birds: the problem of variable hearing abilities. Journal of Field Ornithology 63:26-31.

Evans, J. P., N. Pelkey, and D. G. Haskell. 2002. An assessment of forest change on the Cumberland Plateau in southern Tennessee. Small Area Assessment Forestry Demonstration Project for the Southern Forest Resource Assessment. Report to U.S. Environmental Protection Agency and U.S. Fish and Wildlife Service. Landscape Analysis Lab, Sewanee, Tennessee, USA.

Faanes, C. A., and D. Bystrak. 1981. The role of observer bias in the North American Breeding Bird Survey. Pages 353-359 in C. J. Ralph and J. M. Scott, editors. Estimating numbers of terrestrial birds. Studies in Avian Biology 6.

Flather, C. H., and J. R. Sauer. 1996. Using landscape ecology to test hypotheses about large-scale abundance patterns in migratory birds. Ecology 77:28-35.

Francis, C. M., J. Bart, E. H. Dunn, K. P. Burnham, and C. J. Ralph. 2005. Enhancing the value of the Breeding Bird Survey: reply to Sauer et al. (2005). Journal of Wildlife Management 69:1327-1332.

Hagan, J. M., P. S. McKinley, A. L. Meehan, and S. L. Grove. 1997. Diversity and abundance of landbirds in a northeastern industrial forest. Journal of Wildlife Management 61:718-735.

Hanowski, J. M., and G. J. Niemi. 1995. A comparison of on- and off-road bird counts: Do you need to go off-road to count birds accurately? Journal of Field Ornithology 66:469-483.

Haskell, D. G., J. P. Evans, and N. W. Pelkey. 2006. Depauperate avifauna in plantations compared to forests and exurban areas. PLoS ONE 1: e63. doi:10.1371/journal.pone.0000063. [online] URL: http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0000063.

Heilman, G. E., J. R. Strittholt, N. C. Slosser, and D. A. Dellasala. 2002. Forest fragmentation of the conterminous United States: assessing forest intactness through road density and spatial characteristics. BioScience 52:411-422.

Hunter, W. C., D. A. Beuhler, R. A. Canterbury, J. L. Confer, and P. B. Hamel. 2001. Conservation of disturbance-dependent birds in eastern North America. Wildlife Society Bulletin 29:440-455.

Hutto, R. L., S. J. Hell, J. F. Kelly, and S. M. Pletschet. 1995. A comparison of bird detection rates derived from on-road versus off-road point counts in northern Montana. Pages 103-110 in C. J. Ralph, J. R. Sauer, and S. Droege, editors. Monitoring bird populations by point counts. U.S. Forest Service General Technical Report PSW-GTR-149.

James, F. C., C. E. McCullough, and D. A. Wiedenfield. 1996. New approaches to the analysis of population trends in land birds. Ecology 77:13-27.

Keller, C. M. E., and M. R. Fuller. 1995. Comparison of birds detected from roadside and off-road point counts in Shenandoah National Park. Pages 111-115 in C. J. Ralph, J. R. Sauer, and S. Droege, editors. Monitoring bird populations by point counts. USDA, Forest Service General Technical Report PSW-GTR-149. Albany, California, USA.

Keller, C. M. E., and J. T. Scallan. 1999. Potential roadside biases due to habitat changes along Breeding Bird Survey routes. Condor 101:50-57.

Kendall, W. L., B. C. Peterjohn, and J. R. Sauer. 1996. First-time observer effects in the North American Breeding Bird Survey. Auk 113:823-829.

Koenig, W. D. 1998. Spatial autocorrelation in California land birds. Conservation Biology 12:612-620.

Lawler, J. L., and R. J. O’Connor. 2004. How well do consistently monitored Breeding Bird Survey routes represent the environments of the conterminous United States? Condor 106:801-814.

Link, W. A., and J. R. Sauer. 1994. Estimating equations estimates of trends. Bird Populations 2:23-32.

Link, W. A., and J. R. Sauer. 1998. Estimating population change from count data: application to the North American Breeding Bird Survey. Ecological Applications 8:258-268.

Litvaitis, J. A. 2003. Are pre-Columbian conditions relevant baselines for managed forests in the northeastern United States. Forest Ecology and Management 185:113-126.

McGrath, D. A., J. P. Evans, C. K. Smith, D. G. Haskell, N. W. Pelkey, R. R. Gottfried, C. D. Brockett, M. D. Lane, and E. D. Williams. 2004. Mapping land-use change and monitoring the impacts of hardwood-to-pine conversion on the southern Cumberland Plateau in Tennessee. Earth Interactions 8:1-24.

Meents, J. K., J. Rice, B. W. Anderson, and R. D. Ohmart. 1983. Nonlinear relationships between birds and vegetation. Ecology 64:1022-1027.

O’Connor, R. J., E. Dunn, D. H. Johnson, S. L. Jones, D. Petit, K. Pollock, C. R. Smith, J. L. Trapp, and E. Welling. 2000. A programmatic review of the North American Breeding Bird Survey. Report of a peer review panel. Patuxent Wildlife Research Center, Laurel, Maryland, USA. Available online at: http://www.pwrc.usgs.gov/BBS/bbsreview/bbsfinal.pdf.

Oehler, J. D. 2003. State efforts to promote early-successional habitats on public and private lands in the northeastern United States. Forest Ecology and Management 185:169-177.

Peterjohn, B. G., and J. R. Sauer. 1994. Population trends of woodland birds from the North American Breeding Bird Survey. Wildlife Society Bulletin 22:155-164.

Rich, T. D., C. J. Beardmore, H. Berlanga, P. J. Blancher, M. S. W. Bradstreet, G. S. Butcher, D. W. Demarest, E. H. Dunn, W. C. Hunter, E. E. Iñigo-Elias, J. A. Kennedy, A. M. Martell, A. O. Panjabi, D. N. Pashley, K. V. Rosenberg, C. M. Rustay, J. S. Wendt, and T. C. Will. 2004. Partners in flight North American landbird conservation plan. Cornell Lab of Ornithology, Ithaca, New York, USA.

Robbins, C. S., J. R. Sauer, R. S. Greenberg, and S. Droege. 1989. Population declines in North American birds that migrate to the neotropics. Proceedings of the National Academy of Sciences 86:7658-7662.

Rosenberg, K. V., and P. J. Blancher. 2005. Setting numerical population objectives for priority landbird species. Pages 57-67 in C. J. Ralph and T. D. Rich, editors. Bird conservation implementation and integration in the Americas: Proceedings of the Third International Partners in Flight Conference. U.S. Department of Agriculture Forest Service, General Technical Report PSW-GTR-191.

Rotenberry, J. T., and S. T. Knick. 1995. Evaluation of bias in roadside point count surveys of passerines in shrubsteppe and grassland habitats in southwestern Idaho. Pages 99-101 in C. J. Ralph, J. R. Sauer, and S. Droege, editors. Monitoring bird populations by point counts. U.S. Forest Service General Technical Report PSW-GTR-149.

Royle, J. A., and W. A. Link. 2005. A general class of multinomial mixture models for anuran calling survey data. Ecology 86:2505-2512.

Sauer, J. R., and S. Droege. 1990. Recent population trends of the Eastern Bluebird. Wilson Bulletin 102:239-252.

Sauer, J. R., J. E. Hines, and J. Fallon. 2005a. The North American Breeding Bird Survey, results and analysis 1966–2005. Version 6.2.2006. Patuxent Wildlife Research Center, Laurel, Maryland, USA. Available online at: http://www.mbr-pwrc.usgs.gov/bbs/.

Sauer, J. R., W. A. Link, J. D. Nichols, and J. A. Royle. 2005b. Using the North American Breeding Bird Survey as a tool for conservation: a critique of Bart et al. (2004). Journal of Wildlife Management 69:1321-1326.

Sauer, J. R., W. A. Link, and J. A. Royle. 2004. Estimating population trends with a linear model: technical comments. Condor 106:435-440.

Sauer, J. R., G. W. Pendleton, and B. G. Peterjohn. 1996. Evaluating causes of population change in North American insectivorous songbirds. Conservation Biology 10:465-478.

Sauer, J. R., B. C. Peterjohn, and W. A. Link. 1994. Observer differences in the North American Breeding Bird Survey. Auk 111:50-62.

Scott, J. M., and F. L. Ramsey. 1981. Effects of abundant species on the ability of observers to make accurate counts of birds. Auk 98:610-613.

Sherald, J. L., T. M. Stidham, J. M. Hadidian, and J. E. Hoeldtke. 1996. Progression of the dogwood anthracnose epidemic and the status of flowering dogwood in Catoctin Mountain Park. Plant Disease 80:310-312.

Tennessee Spatial Data Server. 2005. Tennessee land cover. Available online at: http://www.tngis.org/coverages/tnlandcov.zip.

Thogmartin, W. E., A. L. Gallant, M. G. Knutson, T. J. Fox, and M. J. Suárez. 2004. A cautionary tale of the National Land Cover Dataset 1992. Wildlife Society Bulletin 32:960-968.

Thogmartin, W. E., F. P. Howe, F. C. James, D. H. Johnson, E. T. Reed, J. R. Sauer, and F. R. Thompson. 2006. A review of the population estimation approach of the North American Landbird Conservation Plan. Auk 123:892-904.

United States Census Bureau. 2001. Census 2000 TIGER/Line® files (October 2001 release). United States Census Bureau, Washington, D.C., USA.

United States Geological Survey. 2002. Survey routes for the North American Breeding Bird Survey. United States Geological Survey, Washington, D.C., USA. Available online at: http://www.mbrpwrc.usgs.gov/bbs/geographic_information/nabbs02_mis_alb.zip.

United States Geological Survey. 2006. Earth resources observation and science: Thematic Mapper. United States Geological Survey, Washington, D.C., USA. Available online at: http://edc.usgs.gov/products/satellite/tm.html.

United States Geological Survey. 2007. Accuracy assessment of 1992 national land cover data, region 4. United States Geological Survey, Washington, D.C., USA. Available online at: http://landcover.usgs.gov/accuracy/table5.php.

Villard, M. A., G. Merriam, and B. A. Maurer. 1995. Dynamics in subdivided populations of Neotropical migratory birds in a fragmented temperate forest. Ecology 76:27-40.


Address of Correspondent:
David G. Haskell
Department of Biology
735 University Avenue
Sewanee, TN 37383 USA
dhaskell@sewanee.edu

Home | Archives | About | Login | Submissions | Notify | Contact | Search