Geographical and taxonomic biases in research on biodiversity in human‐modified landscapes

Abstract

Biodiversity persistence in human‐modified landscapes is crucial for conservation and maintaining ecosystem services. Studies of biodiversity in landscapes where humans live, work, and extract resources could support defensible policy‐making to manage land‐use. Yet, research should cover relevant regions, and biases in study topics should not lead to gaps in the evidence base. We systematically reviewed the literature of biogeography in human‐modified landscapes published in eight eminent biogeography, conservation, and ecology journals to assess geographical bias among biomes and geopolitical regions and taxonomic bias among species groups. We compared research output per biome to area, biome type, species richness, proportion of transformed land, and the ratio of transformed to protected land. We also compared research output per geopolitical region to area, proportion of transformed land, the ratio of transformed to protected land, and human population density. Research output was distributed unequally among biomes, geopolitical regions, and species groups. Biome type was a clear factor in research bias, and forest biomes were the subject of 87% of papers, while species richness was not generally associated with bias. Conservation in human‐modified landscapes is most important in regions with low protected area coverage, high land conversion, and high pressure from human populations, yet the distribution of published papers did not generally reflect these threats. Seventy‐five percent of studies focused on the Americas and Europe, while Africa and Asia were critically understudied. Taxonomically, plants and invertebrates were the most studied groups; however, research output was not correlated with species richness per group. Protected areas alone will not conserve biodiversity in the long term. Thus, a strong biogeographical evidence base is required to support policies for biodiversity maintenance on human‐modified land. Under‐studied regions and species groups deserve further research to elucidate what, where, and how biodiversity persists in human‐modified landscapes to inform conservation policy and enhance efficacy.

Introduction

Conservation actions should be based on scientific evidence to achieve the best possible outcomes and avoid squandering resources (Sutherland 2004). However, science‐based decision‐making relies on a solid foundation of relevant evidence, often an assemblage of peer‐reviewed studies (Sutherland 2004, Pullin and Knight 2009). Scientists, funding agencies, and publishers hold sway over the composition of the evidence base through their influence on which studies are conducted and published (Lawler et al. 2006). If research interests are misaligned with research needs, gaps in the evidence base could compromise conservation efforts. Thus, it is important to monitor the distribution of published research in comparison to emerging research requirements (Lawler et al. 2006).

We performed such an assessment for an increasingly important field of inquiry—biogeography of human‐modified landscapes—concerned with biodiversity patterns and the processes that maintain them in areas where humans live, work, and extract resources. Though the conservation literature has traditionally focused on large, relatively pristine study sites (Fazey et al. 2005a, Felton et al. 2009), research on the biogeography of human‐modified landscapes is accumulating (see Daily 1999, Daily et al. 2001, Rosenzweig 2003). In line with research needs (Chazdon et al. 2009), these studies seek to define where, what, and how biodiversity persists in human‐modified landscapes; how different aspects of diversity co‐vary; and how human actions drive these patterns.

Protected areas (PAs) are an essential part of the overall conservation strategy, but alone, in the long‐term, they are unlikely to conserve biodiversity for several reasons including constraints of location, size, and configuration (Bengtsson et al. 2003, Brooks et al. 2004, Joppa and Pfaff 2009); ongoing management challenges and outside pressures (Kareiva et al. 2007); and climate change (Loarie et al. 2009). Most compellingly, if species cannot persist beyond PA boundaries, loss of speciation rates and pools of potential immigrants to PAs means that conserving, for example, 10% of the Earth's surface (see Brooks et al. 2004) is likely to result in 90% loss of species (Rosenzweig 2003). Already, humanity has commandeered roughly 40% of Earth's land surface for crops and pastures alone (Millennium Ecosystem Assessment 2005), and demand will escalate for food, fiber, fuel, shelter, space, and freshwater (Tilman 2001). Calls for conservation beyond the boundaries of PAs are not novel (e.g., Leopold 1949), but as humans continue to transform natural ecosystems, conservation efforts in rural villages, logging concessions, pastures, fields, and the like will become increasingly important, not only for conservation's sake but also to sustain valuable ecosystem services (e.g., pollination, decomposition, nutrient cycling) (Tscharntke et al. 2005).

Studies on the biogeography of human‐modified landscapes provide an evidence base to support land‐use planning decisions meant to render human dominated land as amenable as possible to biodiversity. For example, studies demonstrate that agroforestry systems maintain on average >60% of the species richness of primary tropical forests (Bhagwat et al. 2008), oil palm plantations support less forest biodiversity than do other tree crops (Fitzherbert et al. 2008), and scattered remnant trees in fields or pastures help maintain forest diversity (Dunn 2000, Fischer et al. 2010). Yet, to achieve success in supporting policy, the evidence base must cover relevant geographical regions and taxonomic groups and be sufficiently comprehensive; conservation decisions ought to be based on adequate understanding of local biodiversity features and the processes that maintain their viability rather than global generalizations (Svancara et al. 2005). While no topic is likely “over‐studied”, when scientific output is severely biased, “under‐studied” topics could hamper conservation efforts. For instance, research on human‐modified forest ecosystems guides strategies to manage plantations to encourage persistence of forest biodiversity. However, managers have applied the same strategies to plantations embedded in grasslands with dubious efficacy (Pryke and Samways 2003, Lipsey and Hockey 2010).

Unfortunately, biases in the topics that scientists study are common, and the drivers of bias are varied. For example, the distribution of research output among species is uneven (Bonnet et al. 2002, Clark and May 2002, Fazey et al. 2005a, Lawler et al. 2006, Pyšek et al. 2008, Felton et al. 2009, Trimble and van Aarde 2010, Griffiths and Dos Santos 2012). Threatened status, economic importance, or ecological impact drive biases somewhat, but so too, and apparently more so, do personal affinities of scientists, funders, and reviewers toward certain species characteristics that may be unrelated to research needs (Bonnet et al. 2002, Lawler et al. 2006, Trimble and van Aarde 2010). There are also clear geographical biases in ecological research. For instance, the study of invasive species is concentrated in the Americas and Europe with little research conducted in Africa and Asia (Pyšek et al. 2008), a pattern that holds for landscape research, climate change ecology, and conservation biology as a whole (Lawler et al. 2006, Felton et al. 2009, Conrad et al. 2011). If studies of biogeography of human‐modified landscapes are biased towards certain topics, scientists and funding agencies may wish to refocus their efforts to ensure sufficient science is available to support conservation in human‐modified landscapes where it is most needed. Specifically, research on biogeography in human‐modified landscapes should be prioritized in areas (or for species groups) where the topic is little studied despite high threat.

We systematically reviewed the literature on the biogeography of human‐modified landscapes to assess the distribution of research output ecologically, among terrestrial biomes; geopolitically, among UN GeoScheme sub‐regions; and taxonomically, among species groups. The ecological subdivision is important because we expect biodiversity to respond more similarly to land‐use management within than among ecosystem types, while the geopolitical comparison may be more relevant to policy‐makers and funding agencies. We investigated the relationship between the distribution of research and the area and species richness per region and biome type. We also compared the distribution of research to the estimated importance of countryside biogeography in each region based on population density, proportion of land converted, and the ratio of land area converted to area protected. These metrics provide a rough quantification of the risk of biome‐wide biodiversity loss and, thus, the importance of conservation outside PAs. We also assessed the distribution of research among seven species groups: birds, fish, fungi, herpetofauna, mammals, plants, and invertebrates.

Methods

Literature search

We searched the ISI Web of Knowledge (covering 1950–2010) in May 2012 with keywords “biodiversity” or “conservation” and each of the following terms: “agricultur*”, “agroforest*”, “crop$land”, “farm*”, “forestry”, “human$modified”, “multiple$use management”, “range$land”, “rural”, “sub$urban”, and “urban” (“*” is a wildcard indicating any group of characters, and “$” represents zero or one character). We limited our assessment to eight conservation, biogeography, and ecology journals: Biodiversity and Conservation, Biological Conservation, Conservation Biology, Diversity and Distributions, Journal of Applied Ecology, Journal of Biogeography, Ecological Applications, and Ecology. From the initial search, we retained primary research papers that assessed occurrence or persistence of multi‐species assemblages on human‐modified land under current use. Thus, we excluded studies of abandoned landscapes, restoration projects that excluded human use, and fragmentation studies that did not explicitly consider biodiversity in the human‐modified areas surrounding the fragments.

The choice of keywords and journals searched was a compromise between practicality and comprehensiveness. Our keywords were, by necessity, not overly specific to avoid biasing search results. Thus, our keywords returned many papers not relevant to the topic, and practicality dictated that we limit the number of journals searched to prevent an unwieldy number of search results. We selected eight journals that we expected to be among the least biased towards particular biome types, regions, or taxonomic groups.

To assess how research output on the topic of biogeography of human‐modified landscapes has changed over time in these journals, we calculated the number of papers per year identified in our literature search, and we noted the total number of papers published per year available on the ISI Web of Knowledge for each journal. We used linear regression to assess changes over time in the proportion of the total research output that was composed of studies on the biogeography of human‐modified landscapes.

Geographical distribution of research output

We assessed the geographical distribution of research output both politically and ecologically. For each paper identified by our search, we noted the geopolitical region or regions where the study was carried out based on an intermediate scale of subdivision, the UN GeoScheme categorization (UNSD 2011; Micronesia, Melanesia, and Polynesia combined to yield 19 geopolitical regions). To assess the ecological distribution, we noted the terrestrial biome or biomes in which research was conducted (Olson et al. 2001, WWF 2001). In the few cases where studies assessed biodiversity in aquatic ecosystems, we allotted terrestrial biomes based on the location of the water bodies.

Many papers identified in the literature search, especially of European origin, considered “farmland biodiversity” in semi‐natural landscapes with no reference to any natural ecosystem. These studies may not represent useful information on conserving the biodiversity of the original biome; for example, studies of biodiversity in extensive semi‐natural grasslands under varying management regimes may or may not inform the conservation of the biodiversity of boreal forests in which the grasslands are embedded. Thus, to assess the effect of the inclusion of such studies on further analyses, in addition to recording the biome in which they took place, we also categorized them as “no comparison” (in contrast to “natural comparison”). “Natural comparison” studies compared biodiversity patterns or processes to those of a natural baseline, either analytically or conceptually, and “no comparison” studies did not.

We expected the area of regions to determine the distribution of research among biomes and geopolitical units if research efforts were randomly distributed geographically. The area covered by the largest biome, deserts and xeric shrublands, eclipses the smallest in our study, tropical and subtropical coniferous forests, by 39 times. Similarly, the largest geopolitical region, Northern America, is 87 times the area of the Caribbean. Thus, we regressed the number of papers per region on area of biomes (Olson et al. 2001, WWF 2001) and geopolitical regions (UNEP 2011a). We then calculated area‐corrected estimates of research output as the number of studies per million km² for biomes and geopolitical regions to investigate whether other factors were related to bias in research output.

For biomes, these factors included biome type (i.e., forest or other), species richness per biome, and the estimated importance of research in a biome. We compared the area‐corrected estimates of research output between the seven forest biomes (i.e., three tropical, two temperate, boreal, and Mediterranean forests) and the six other biomes (i.e., montane, flooded, tropical, and temperate grasslands; tundra; and deserts). To evaluate whether research output was skewed towards biomes with higher species richness, we used Spearman rank correlation to compare the total studies per biome to the estimated total number of mammal, bird, reptile, and amphibian species per biome (Hassan et al. 2005). We also compared area‐corrected studies per biome to an estimate of the biome‐specific z‐value from the power model of the species‐area relationship for vascular plants calculated by Kier et al. (2005) (we averaged the four sub‐regional z‐values for the tropical and subtropical dry broadleaf forests). We reasoned that biogeography of human‐modified landscapes should be most important in biomes that have been heavily transformed and especially in those that also have low PA coverage. Therefore, we used Spearman rank correlation to compare area‐corrected research output to the proportion of transformed land per biome and to the ratio of converted to protected land (i.e., the Conservation Risk Index (CRI) calculated by Hoekstra et al. (2005)). We then used Mann‐Whitney U tests to compare the two species richness metrics, CRI, and proportion of transformed land between forest and other biomes.

For geopolitical regions, we used Spearman rank correlation to assess whether area‐corrected research output was correlated with the proportion of agricultural land conversion (World Bank 2009), the geopolitical CRI (the ratio of land conversion to PA coverage (UNSD 2010)), and population density (UNEP 2011b). We reasoned that conservation beyond PAs would be both particularly important and challenging in geopolitical regions with high population density.

Distribution of research output among species groups

We categorized the species group or groups assessed in each paper as birds, fish, fungi, herpetofauna, mammals, plants, or invertebrates. We then calculated the percent of total, “natural comparison”, and “no comparison” studies that assessed each group. We used Spearman rank correlation to assess whether the proportion of all studies that assessed each species group was correlated with the proportion of described species comprised by each group (Vie et al. 2009).

Results

We assessed the distribution of research output over time, by geo‐ecological region (i.e., biome), by geopolitical region, and by taxonomic group with respect to a number of potential explanatory variables summarized in Table 1 and discussed below.

Summary of variables considered and outcomes of statistical tests with respect to research output groupings of studies of biogeography in human‐modified landscapes published in eight major journals.

Of the 2521 references returned by our literature search, 681 assessed the occurrence and/or persistence of biodiversity in human‐modified landscapes and met our inclusion criteria. These papers (published between 1984 and 2010) made up 4–7% of the total papers published over the same period in Biological Conservation (214 of 4890), Biodiversity and Conservation (168 of 2685), and Journal of Applied Ecology (110 of 2750). They comprised 1.5–2% of research output in Ecological Applications (64 of 2796), Diversity and Distributions (16 of 657), and Conservation Biology (77 of 4132), while they were less prevalent in Ecology (12 of 7385) and Journal of Biogeography (20 of 2837). Prevalence has increased over time, even after accounting for the increase in publishing output (Fig. 1). The yearly proportion of the total studies published by all eight journals comprised of studies identified in our literature review increased significantly with year (F_1,18 = 219.1, p < 0.001, r² = 0.90, y = 0.002x − 3.892) although the pattern was clearly non‐linear over the study period (Fig. 1).

Increase in studies of biogeography of human‐modified landscapes over time. For the eight journals we considered, the proportion of the total studies published comprised by studies of the biogeography of human‐modified landscapes identified in our review increased significantly over time.

We distinguished 218 papers as “no comparison” studies and 463 as “natural comparison”. We performed subsequent analyses both including and excluding “no comparison” studies. Seven percent of the total studies and 8% of “natural comparison” studies considered biodiversity of human‐modified aquatic ecosystems, e.g., streams, wetlands, and ponds.

Geographical bias

The number of studies per biome (see Fig. 2) differed widely from one in tundra to 316 in temperate broadleaf and mixed forests, while tropical and subtropical moist broadleaf forests had the highest number of studies from the “natural comparison” category (169) (Fig. 3). The seven forest biomes were the subject of 87% of studies. Contrastingly, only 13% of papers assessed the other six biomes. While “no comparison” studies came almost exclusively from forest biomes (96%), excluding them did not remove the bias towards forests; 83% of “natural comparison” studies were conducted in forest biomes.

World map of coverage of 14 terrestrial biomes. The 14 terrestrial biomes adapted from Olson et al. (2001).

Distribution of research output, CRI, and proportion of land transformed among biomes. Discrepancy between the number of biogeography of human‐modified landscape studies per million km2 (dark green = “natural comparison” studies; light green = “no comparison” studies; total number of studies listed to the right of bars) and Conservation Risk Index (CRI, blue bars) per terrestrial biome. Per‐biome proportion of land that is transformed is listed on the right‐hand axis. Biome abbreviations: Boreal for./taiga = Boreal forests/taiga; Montane g./sh. = Montane grasslands and shrublands; Temp. Con. For. = Temperate conifer forests; Deserts/x. sh. = Deserts and xeric shrublands; Flooded g./sav. = Flooded grasslands and savannas; Trop./sub. g./sav./sh. = Tropical and subtropical grasslands, savannas, and shrublands; Trop./sub. moist br. for. = Tropical and subtropical moist broadleaf forests; Trop./sub. con. for. = Tropical and subtropical coniferous forests; Temp. br./mix. for. = Temperate broadleaf and mixed forests; Trop./sub. dry br. for. = Tropical and subtropical dry broadleaf forests; Med. for./wd./scrub = Mediterranean forests, woodlands, and scrub; Temp. g./sav./sh. = Temperate grasslands, savannas, and shrublands.

Research output was also uneven among geopolitical regions (Figs. 4 and 5), ranging from zero studies in Melanesia, Micronesia, and Polynesia to 163 in Northern Europe. Forty‐two percent of the papers studied European regions (although 68% of these were “no comparison” studies), while a further 33% centered on regions in the Americas. Studies conducted in Australia and New Zealand, Africa, and Asia each made up <10% of the total studies.

Distribution of research output, CRI, and proportion of land transformed among geopolitical regions. Discrepancy between number of biogeography of human‐modified landscape studies per million km2 (dark green = “natural comparison” studies; light green = “no comparison” studies; total number of studies listed to the right of bars) and Conservation Risk Index (CRI, blue bars) per geopolitical region. Per‐region proportion of land that is transformed is listed on the right‐hand axis. Geopolitical region abbreviations: Mel./Micro./Poly. = Melanesia, Micronesia, and Polynesia; Aust./N.Z. = Australia and New Zealand.

World map of CRI and research output per geopolitical region. Geopolitical regions based on UN GeoScheme (UNSD 2011) colored according to their Conservation Risk Index (CRI) from low (yellow) to high (red); blue circles are proportional in size to the area corrected research output per geopolitical region (refer to Fig. 4 for values).

Research output was not randomly distributed among biomes or geopolitical regions. The number of studies per biome and geopolitical region were not related to area (biomes: F_1,11 = 1.87, p = 0.20; geopolitical regions: F_1,18 = 0.07, p = 0.80), even when “no comparison” studies were excluded (biomes: F_1,11 = 2.34, p = 0.15; geopolitical regions: F_1,18 = 2.40, p = 0.14).

The median number of studies per million km² of forest biomes (8.65) was much higher than for other biomes (1.49) (Mann‐Whitney U = 1.00, p < 0.01) (Fig. 6). The difference remained substantial and significant even when “no comparison” studies were excluded (median studies in forest biomes = 8.46, other biomes = 1.49, Mann‐Whitney U = 2.00, p < 0.01). If we categorized Mediterranean forests, woodlands, and scrub as “other” instead of “forest”, research output remained biased towards forest in the total dataset (median studies in forest biomes = 8.55, other biomes = 1.54, Mann‐Whitney U = 6.00, p = 0.04) and when “no comparison” studies were excluded (median studies in forest biomes = 7.16, other biomes = 1.54, Mann‐Whitney U = 7.00, p = 0.05).

Research output per biome type. Median number of studies per million km2 in forest biomes was significantly higher than in other biomes both including and excluding “no comparison” studies.

The number of studies per biome was not significantly correlated with estimated total mammal, bird, reptile, and amphibian richness including “no comparison” studies (Spearman r = 0.14, p = 0.65) or excluding them (Spearman r = 0.31, p = 0.30). However, the correlation between studies per million km² and biome‐specific z‐values for vascular plants was significant when “no comparison” studies were excluded (Spearman r = 0.56, p = 0.05), but not when they were included (Spearman r = 0.51, p = 0.08). However, z‐values were higher in forest than other biomes (Mann‐Whitney U = 5.50, p = 0.03), while total species of mammals, birds, reptiles, and amphibians did not differ significantly (Mann‐Whitney U = 18.00, p = 0.37).

Area‐corrected research output was not correlated with the per‐biome CRI (Spearman r = 0.55, p = 0.051) unless “no comparison” studies were excluded (Spearman r = 0.63, p = 0.02) (Fig. 3). However, research output was correlated with proportion of land converted per biome both with and without “no comparison” studies (Spearman r = 0.65, p = 0.02; Spearman r = 0.73, p < 0.01). However, CRI did not differ significantly between forest biomes and others (Mann‐Whitney U = 15.00, p = 0.43), nor did proportion of land converted (Mann‐Whitney U = 12.00, p = 0.23). Additionally, area‐corrected research output was not correlated with CRI per geopolitical region (all data: Spearman r = −0.18, p = 0.22; “no comparison” studies excluded: Spearman r = −0.15, p = 0.26), nor with proportion of land converted to agriculture (all data: Spearman r = 0.11, p = 0.32; “no comparison” studies excluded: Spearman r = 0.14, p = 0.27) (Fig. 4). However, area‐corrected research output per geopolitical region was weakly correlated with population density (all data: Spearman r = 0.47, p = 0.04; “no comparison” studies excluded: Spearman r = 0.47, p = 0.03).

Taxonomic bias

Research output was not distributed evenly among seven major taxonomic groups (Fig. 7). Of the 681 papers indentified (13% of which studied multiple species groups), 36% assessed invertebrates. Birds and plants were each assessed in 31% of papers. Contrastingly, fewer studies assessed mammals (10%), herpetofauna (7%), fungi (3%), and fish (0.5%).

Distribution of research output and estimated richness per taxonomic group. Discrepancy between the proportion of the 681 biogeography of human‐modified landscape studies that assessed each taxonomic group (dark green = “natural comparison” studies; light green = “no comparison” studies) and the estimated proportion of richness per group (blue bars).

The 218 “no comparison” studies focused even more on invertebrates (47%) and plants (39%). Of “no comparison” studies, 25% assessed birds, while only 3%, 2%, 1%, and 0% covered mammals, herpetofauna, fungi, and fish respectively. Therefore, among the 463 “natural comparison” studies, research output was more evenly distributed among taxonomic groups (Fig. 7): birds (33%), invertebrates (31%), plants (27%), mammals (13%), herpetofauna (9%), fungi (4%), and fish (1%). However, the proportion of studies that assessed each group was not correlated with the proportion of described species per group (all data: Spearman r = 0.34, p = 0.44; “no comparison” studies: Spearman r = 0.40, p = 0.40; “natural comparison” studies: Spearman r = 0.11, p = 0.84).

Discussion

Biogeography of human‐modified landscapes provides the evidence base required to support defensible policy‐making to encourage biodiversity conservation beyond protected areas, an increasingly important objective. We have shown that it has been a growing sub‐discipline in conservation biology over the past two decades, as reflected by publication trends in the eight journals included in our assessment. We have also demonstrated, however, that scientific research output is biased geo‐ecologically, geopolitically, and taxonomically. Geo‐ecologically, research output for forest biomes was disproportionately higher than for other biomes after correcting for area. In particular, temperate broadleaf and mixed forests and tropical and subtropical moist broadleaf forests garnered a large proportion of research output. Geopolitically, the bias was clearly towards Europe and the Americas, while substantially fewer studies came from Africa and Asia. Taxonomically, research attention among species groups was neither evenly distributed nor correlated with per‐group richness, and invertebrates, plants, and birds were the most studied groups.

Our literature search was extensive, covering 681 papers, though not comprehensive. We searched eight journals for a limited set of search terms because practicality dictated that we could not assess all papers ever published. We attempted to minimize bias as far as possible in our selection of relatively neutral keywords and journals. Additionally, the journals we selected are preeminent in conservation, ecology, and biogeography, and we expect them to be representative of the wider research base of high‐quality studies available to and useful for policy‐makers. Nonetheless, further consideration of less prestigious journals, the grey literature, and non‐English language publications may influence the conclusions of this study.

“Natural comparison” versus “no comparison” studies

We identified two types of studies: “natural comparison” and “no comparison”. “Natural comparison” studies compared biodiversity between human land‐use and a baseline reference from relatively natural fragments of the biome in which the study was conducted. For example, many studies in tropical forests compared biodiversity among agroforestry plantations, crop fields, pastures, and nearby PAs or forest remnants (e.g., Zapfack et al. 2002, Wanger et al. 2010).

On the other hand, “no comparison” studies, which came predominantly from Europe, lacked direct reference to natural ecosystems. Many assessed the effects of agricultural management (e.g., organic versus conventional (e.g., Batary et al. 2010) or an intensification gradient (e.g., Kohler et al. 2007)) on “farmland biodiversity”, the suite of species occupying traditionally managed agro‐ecosystems (see Bignal and McCracken 1996), without specific reference to the biomes in which the studies were conducted. Often, farmland itself is presented as a novel ecosystem worthy of conservation for its own sake (e.g., Stefanescu et al. 2005, Jay‐Robert et al. 2008), best accomplished by promoting the traditional agricultural practices that created it over thousands of years (e.g., Bignal and McCracken 1996, Pykala 2000). For example, semi‐natural grasslands created by traditional agricultural practices in Europe's forest biomes are particularly important in conservation schemes (see Austrheim et al. 1999, Walker et al. 2004), and many papers compared biodiversity among varying management options for maintaining them (e.g., Poyry et al. 2005, Saarinen and Jantunen 2005).

Although such studies provide crucial support for conservation in Europe, their applicability elsewhere and relevance to the biome in which they were conducted cannot be assumed. Thus, we distinguished these studies in our analyses of research output bias and associated factors. Nonetheless, the recognized value of farmland biodiversity in landscapes long dominated by humans (see Bignal and McCracken 1996, Pykala 2000) testifies to the importance of the early consideration of biogeography of human‐modified landscapes in land‐use planning for regions that retain large tracts of relatively undisturbed land (e.g., wilderness areas (Mittermeier 2003)). A comparative approach is important to inform conservation strategies in human‐modified landscapes because it allows for consideration of community composition and functional trait richness over space and time (e.g., Mayfield et al. 2006, Flynn et al. 2009), investigation of processes that link occurrence to persistence (e.g., Trimble and van Aarde 2011), and, to avoid biotic homogenization, distinction between land‐uses amenable to invasive or cosmopolitan species versus more localized species (see Filippi‐Codaccioni et al. 2010).

Patterns of geographical bias in research output

We expected per‐region area to determine the distribution of research output if research interest were distributed randomly geographically. However, this was not the case for biomes or geopolitical regions. For biomes, we investigated ecosystem type (forest or other), per‐biome species richness, the proportion of transformed land, and CRI in relation to research distribution.

Area‐corrected research output was clearly higher in forest than other biomes, a pattern also reported by Fazey et al. (2005a) in the conservation literature. Felton et al. (2009) subsequently found twice the degree of bias towards forests (38% of studies versus 20.5%) in the climate‐change ecology literature. We found more than double that again (87% of studies) for the literature on biogeography of human‐modified landscapes. The area‐corrected number of studies per biome was positively correlated with the z‐value from the power model of the species‐area relationship for vascular plants (Kier et al. 2005) when “no comparison” studies were excluded. However, z‐values were also significantly higher in forest biomes than other types. On the other hand, the per‐biome number of mammal, bird, reptile, and amphibian species was not correlated with the number of studies per biome and did not differ significantly between forest and other types. Additionally, the overall bias towards forests could not be explained by an estimated increased importance of study there because forests did not tend to have a higher CRI or proportion of transformed land than did other biomes. Thus, while the high plant richness of forests may play some role in research bias, it seems a penchant for forests, rather than species richness or threat status per se, drives bias. Felton et al. (2009) suggested that the bias towards forests in the climate‐change ecology literature was a result of the concomitant bias towards North American and European study sites. However, our results indicate that the forest bias is prominent in both temperate and tropical regions.

While research output per biome was correlated, based on rank, with CRI when “no comparison” studies were excluded and with proportion of transformed land, there did seem to be under‐studied biomes. The temperate grasslands, savannas, and shrublands biome was the most glaring example, and tropical and subtropical dry broadleaf forests warranted more attention. The Mediterranean forests, woodlands, and scrub biome had a relatively high number of studies after correcting for area, but its CRI and the proportion of land converted were very high, warranting more research. Among the biomes on the lower end of the CRI scale, there was a relative excess of studies from the temperate conifer forests and the boreal forests, which both have low proportions of converted land. Contrastingly, tropical and subtropical grasslands, savannas, and shrublands; flooded grasslands and savannas; and deserts and xeric shrublands were under‐studied relative to their CRI.

Geopolitical bias was towards Europe and the Americas with far less focus on Africa and Asia, a pattern previously demonstrated for other sub‐disciplines of conservation and ecology (Pyšek et al. 2008, Felton et al. 2009, Conrad et al. 2011). Disconcertingly, area‐corrected research output per geopolitical region was not correlated with CRI. Central Asia, Eastern Asia, Southern Asia, Northern Africa, and West Africa had a particularly low research output considering their high CRI. Additionally, while research output per region was weakly correlated with human population density, several regions with high population densities had low research output including Southern, Eastern, and South‐eastern Asia and the Caribbean.

CRI, proportion of land converted, and population density are indices that we expect to highlight regions where conservation beyond PAs will be especially important due to extensive land conversion, little protection, and high threat. The general disparities in patterns of research output relative to these estimates of research importance could act as a guide to where additional research investment in the biogeography of human‐modified landscapes may be most beneficial. There are, however, caveats to consider. Estimates of land conversion do not account for the likelihood of future conversion or intensification, and they likely underestimated land‐uses that did not totally transform the land, yet might result in substantial degradation, e.g., grazed rangelands, hunting grounds, and selectively logged forests (Hoekstra et al. 2005, World Bank 2009).

Additionally, the focus of this paper was terrestrial, but marine and aquatic ecosystems also warrant consideration. While 7% of studies from our search assessed biodiversity in human‐modified ponds, streams, and wetlands, it was not feasible to include a more targeted search in this assessment. However, we predict that similar knowledge gaps exist regarding biogeography of human‐modified aquatic biomes, an issue for future study.

Patterns of taxonomic bias

Unsurprisingly, research output was distributed unequally among seven taxonomic groups and was not correlated with per‐group richness. However, the disparities did not mirror those found in other studies. We found that plants, invertebrates, and birds received the most attention, while mammals, herpetofauna, fish, and fungi were studied much less often. Contrastingly, Clark and May (2002) demonstrated that in the general conservation literature, vertebrate species command the attention of 69% of papers, while invertebrates (11%) and plants (20%) get less attention even though vertebrates comprise a small fraction of known species (3%) compared to invertebrates (79%) and plants (18%).

Also in contrast to our findings, other studies have demonstrated a strong bias towards mammals compared to other vertebrates (Bonnet et al. 2002, Clark and May 2002, Trimble and van Aarde 2010). In a survey of papers published in nine leading ecological journals, Bonnet et al. (2002) found that birds and mammals are over represented compared to their species diversity (with 44% and 27% of papers respectively yet only 20% and 9% of vertebrate species), while fish (14% of papers, 48% of species), reptiles (7% of papers, 14% of species), and amphibians (7% of papers, 9% of species) are underrepresented.

Therefore, within the field of biogeography of human‐modified landscapes, research output among taxonomic groups may more closely mirror species richness among groups than does the general ecological or conservation literature. However, this reduced bias compared to other fields may be better explained by methodological constraints rather than a sense of fairness among researchers (Pawar 2003). Invertebrates, plants, and birds may be easier to survey than mammals, herpetofauna, and fish. The former groups may also be more likely than the latter to persist in and, thus, be available for study in human‐modified landscapes.

Yet, species richness per taxonomic group may not be the best resource allocation metric for research. Although one aim of promoting biodiversity persistence in human‐modified landscapes is to complement PAs in conserving species, the other is to maintain ecosystem services (Chazdon et al. 2009). That “no comparison” studies were more biased towards plants and invertebrates than “natural comparison” studies were is perhaps a reflection that these groups are most closely linked with ecosystem services, such as pollination and pest control, upon which agriculture relies (e.g., Albrecht et al. 2007, Bell et al. 2008).

Origins of bias and new directions for research

Although biome type, geopolitical region, and species group were related to bias, definitively determining the root cause of the biases in research output from the eight journals assessed here was beyond the scope of this paper. Research history and interests of individuals and organizations, priorities of funding agencies and governments, and the stance of journal editorial boards are all likely to play a role in influencing the type of research conducted and published (see Fazey et al. 2005b). So too are practical constraints such as the varying difficulty of conducting research in different geographical regions, varying support and capacity for science in different countries, and language barriers to publication (see Fazey et al. 2005b, Griffiths and Dos Santos 2012), in conjunction with regional disparities in economic incentives and resources (see Pyšek et al. 2008), to name just a few.

This systematic review has highlighted gaps in the evidence base of biogeography of human‐modified landscapes that scientists, publishers, funding agencies, and governments may wish to consider when planning future research and making decisions that affect research output. While research output among taxonomic groups was not free from bias, we conclude that the geo‐ecological and geopolitical biases are more immediate hurdles for science‐based conservation action (see also Pyšek et al. 2008). Some biomes have attracted a good deal of research interest (particularly temperate broadleaf and mixed forests and the tropical and subtropical moist broadleaf forests), while other biomes were critically under‐investigated. Similarly, research output was biased towards geopolitical regions in Europe and the Americas, yet Asian and African regions were generally severely underrepresented.

As conservation efforts beyond PAs become increasingly important globally, these deficiencies could have profound consequences. Conservation success on human‐modified land depends on a sound and comprehensive evidence base and interdisciplinary collaboration to meet humanity's demands for resources while allowing the persistence of biodiversity. The evidence base to support sensible land‐use policies needs to have been generated in an appropriate geo‐ecological and geopolitical context and be extensive enough to allow systematic review or meta‐analytical assessment to draw robust conclusions regarding management actions (Pullin and Knight 2009, Segan et al. 2011, Guldemond et al. 2012).

We hope this paper will be the first step towards rectifying the gaps in the evidence base that we have identified. Ideally, awareness of the current biases will lead researchers, funding agencies, editors, and publishers to choose, of their own volition, to prioritize biogeography of human‐modified landscapes in under‐studied regions and biomes, while continuing to develop and refine research and implementation strategies for the regions that have already attracted a good deal of research. Obviously, international funding agencies could do a great deal to support research in under‐studied regions. Similarly, government policies and funding opportunities that encourage international scientific collaboration could help spread resources to under‐studied regions, promote valuable knowledge exchange, and build local capacity (Fazey et al. 2005b). However, given that similar research biases have emerged repeatedly in the conservation and ecology literature (e.g., Fazey et al. 2005b, Lawler et al. 2006, Pyšek et al. 2008, Felton et al. 2009, Conrad et al. 2011), and little progress has been made (Griffiths and Dos Santos 2012), future work should specifically assess how to encourage research on topics in need of more attention.

Acknowledgments

M.J.T. is supported by a National Science Foundation Graduate Research Fellowship and R.J.v.A. through various grants to the Chair in Conservation Ecology at CERU.