Evaluation of Museum Collection Data for Use in Biodiversity Assessment

Evaluation of Museum Collection Data for Use in Biodiversity Assessment W. F. PONDER,* G. A. CARTER, P. FLEMONS, AND R. R. CHAPMAN Centre for Biodiver...
Author: Evelyn May
0 downloads 0 Views 607KB Size
Evaluation of Museum Collection Data for Use in Biodiversity Assessment W. F. PONDER,* G. A. CARTER, P. FLEMONS, AND R. R. CHAPMAN Centre for Biodiversity and Conservation Research, Australian Museum, 6 College Street, Sydney, NSW 2010, Australia

Abstract: Natural-history collections in museums contain data critical to decisions in biodiversity conservation. Collectively, these specimen-based data describe the distributions of known taxa in time and space. As the most comprehensive, reliable source of knowledge for most described species, these records are potentially available to answer a wide range of conservation and research questions. Nevertheless, these data have shortcomings, notably geographic gaps, resulting mainly from the ad hoc nature of collecting effort. This problem has been frequently cited but rarely addressed in a systematic manner. We have developed a methodology to evaluate museum collection data, in particular the reliability of distributional data for narrow-range taxa. We included only those taxa for which there were an appropriate number of records, expert verification of identifications, and acceptable locality accuracy. First, we compared the available data for the taxon of interest to the “background data,” comprised of records for those organisms likely to be captured by the same methods or by the same collectors as the taxon of interest. The “adequacy”of background sampling effort was assessed through calculation of statistics describing the separation, density, and clustering of points, and through generation of a sampling density contour surface. Geographical information systems (GIS) technology was then used to model predicted distributions of species based on abiotic (e.g., climatic and geological) data. The robustness of these predicted distributions can be tested iteratively or by bootstrapping. Together, these methods provide an objective means to assess the likelihood of the distributions obtained from museum collection records representing true distributions. Potentially, they could be used to evaluate any point data to be collated in species maps, biodiversity assessment, or similar applications requiring distributional information. Análisis del Uso de Datos de Colecciones de Museo en la Evaluación de la Biodiversidad Resumen: Las colecciones de historia natural en museos contienen datos críticos para la toma de decisiones en la conservación de la biodiversidad. Colectivamente estos datos basados en especimenes describen las distribuciones de taxa conocidos en tiempo y espacio. Como las más confiables fuentes integrales de información para la mayoría de las especies descritas, estos registros son potencialmente accesibles como una solución de costo-utilidad a un amplio rango de preguntas de conservación e investigación. No obstante, estos datos tienen deficiencias, brechas geográficas notables, resultando principalmente de la naturaleza ad hoc de los esfuerzos de colecta. Este problema ha sido citado frecuentemente pero raramente ha sido abordado de una manera sistemática. Desarrollamos una metodología para evaluar datos de colecciones de museo, en particular la confiabilidad de los datos de distribución de taxones de rango corto. Solo se incluyeron aquellos datos para los que existió un número adecuado de registro, verificaciones expertas de identificaciones y una aceptable precisión de la localidad. Los datos disponibles para los taxones de interés fueron comparados con “datos previos” compuestos de registros de aquellos organismos propensos a ser capturados por los mismos métodos y por los mismos colectores de los taxones de interés. La “adecuación” de los esfuerzos de muestreos previos fue evaluada mediante cálculos estadísticos que describieran la separación, la densidad y el anidado de los puntos y la generación de un muestreo de un contorno de superficie de la densidad. Se empleó tecnología de Sistemas de información Geográfica (GIS) para modelar las distribuciones predichas de las especies basándose en datos abióticos (climáticos y geológicos). La contundencia de estas distribu*email [email protected] Paper submitted November 4, 1999; revised manuscript accepted July 19, 2000.

648 Conservation Biology, Pages 648–657 Volume 15, No. 3, June 2001

Ponder et al.

Evaluating Museum Data

649

ciones predichas puede ser evaluada iterativamente o por el método de “bootstrapping”. En conjunto, estos métodos proveen medios objetivos para evaluar la posibilidad de que las distribuciones obtenidas de registros de colecciones de museo representen las distribuciones verdaderas. Esto podría ser usado para evaluar cualquier dato a ser cotejado en mapas de especies, evaluaciones de biodiversidad, o aplicaciones similares que requieren información de distribución.

Introduction The value of museum natural-history collections for biodiversity research has been recently emphasized (e.g., Hoagland 1989; Hawksworth & Mound 1991; Nielsen & West 1994; Davis 1996 and references therein). These collections are essentially huge databases that have accumulated over long periods and that can thus provide a historical perspective to complement contemporary field surveys. In contrast to other forms of data, they are regularly accessed by specialists, and the material they contain is available for validation in the future, making them essential to the integrity of biological knowledge (Cotterill 1995, 1997). Museum collections may be particularly useful as a source of data on invertebrates, which comprise around 99% of animal life but have been largely ignored in most conservation and management programs. Although the need to consider invertebrates has gained considerable recognition in the last few years (e.g., Disney 1986; Majer 1987; Kremen et al. 1993; Heywood 1995; New 1998; Ponder & Lunney 1999), it is often considered difficult because of the relative lack of published data, perceived time and cost involved in wide-ranging invertebrate surveys, and lack of appropriate taxonomic expertise. Museum collections hold detailed distributional data for many groups of invertebrates, and it is important that these data be evaluated and, where adequate, utilized in conservation planning. Problems with the use of collection data include lack of access or availability, as well as inherent shortcomings (especially geographic gaps) in the data themselves. Often the data are unpublished or the collections are not electronically databased, and collections are usually arranged taxonomically, necessitating inefficient manual retrieval of details for large suites of taxa. Shortcomings of the data include the ad hoc nature of the collections, presence-only data, biased sampling, and large collecting gaps in time and space, although the extent of such gaps depends on both the area and group under study. For instance, although the British biota has been extremely well sampled (e.g., Prendergast et al. 1993a), a study on Amazonian taxa (Kress et al. 1998) found that more than a quarter of the primary grid cells lacked data. Some taxa, such as the small flies discussed by Bickel (1999), require specialized collecting and are less likely to be represented in invertebrate collections than more con-

spicuous groups such as butterflies, larger beetles, large land snails, or freshwater crayfish. Collection data have been used in a number of recent papers to map species distributions or determine areas of conservation importance (e.g., Prendergast et al. 1993a; Väisänen & Heliövaara 1994; Vane-Wright et al. 1994; Drinkrow & Cherry 1995; Skelton et al. 1995; Williams et al. 1996; Kress et al. 1998; McCarthy 1998). But there have been relatively few attempts to address the shortcomings of collection data. The problem of undersampling during inventories was addressed by Colwell and Coddington (1994), who used several techniques designed to give a measure of survey completeness. Nelson et al. (1990) showed that supposed centers of endemism were strongly correlated with collecting intensity, whereas Prendergast et al. (1993b) and Fagan and Kareiva (1997) introduced methods to correct for sampling effort in determining the number of species in a region. The mapping of collection records for individual taxa may allow some checking of the spatial accuracy of the data (e.g., the position of apparent outliers), but it gives little indication as to whether the collection records accurately reflect the true distribution of the taxon. We developed a methodology to determine the reliability of distributional data, with a particular focus on seemingly narrow-range taxa. Species represented by one or a few records may be widespread but “rare” (only occasionally found, despite searching effort) or simply infrequently collected through lack of effort. The problem is therefore to distinguish between these taxa and those that can be reliably judged to be of very restricted distribution. Our methods were developed as part of the Australian Museum’s Narrow Range Endemics Program (NREP) (Ponder 1999), the aim of which is to identify areas important for the conservation of narrow-range, nonmarine invertebrate taxa (i.e., endemicity “hotspots”; Myers 1988, 1990). The approach is flexible and could potentially be used to evaluate any distributional data obtained from museum or herbarium collections or other sources.

Methods We evaluated the robustness of distributions mapped from collection data in two stages. In step 1 we compared the available data for each species with the “background data,” an indication of background sampling ef-

Conservation Biology Volume 15, No. 3, June 2001

650

Evaluating Museum Data

fort or sampling intensity. In step 2 we used geographic information system (GIS) technology for predictive species modeling. Data Selection The data had to meet a number of criteria to be included in the analysis. “Taxonomic groups” (genera, families, or higher groups) were acceptable providing that they (1) had been sampled at an appropriate intensity in the area of interest and (2) contained one or more “taxa of interest” or were suitable as part of a “background group” for a particular taxon of interest. “Taxon of interest” (TI) refers to the particular species-group taxon (usually a species) whose distribution is being analyzed. The choice of TI depends primarily on the aim of the analysis: in the NREP, for example, the TIs are all narrow-range taxa. The number of records (samples from a particular spot on a particular occasion) available for each TI is also important, however. The estimation of distributions of taxa known only from a small number of collection events can be equivocal, especially if the taxa are spread over a relatively large area or if background sampling is sparse. We used a minimum of three records, but this could be increased to improve reliability. The protocols we used for the inclusion of individual “records” were as follows: (1) the same expert verifies the identity of all records accepted for a particular TI; unnamed taxa (e.g., morphospecies) were also acceptable and were coded; and (2) the locality accuracy is appropriate to the scale of the particular question being addressed. In the NREP, all records are assigned a six-category accuracy code (10 m, 10–100 m, 100 m–1 km, 1–10 km, 10–100 km, 100 km) and can be filtered accordingly. Records can also be sorted by date, so that, for example, records older than a defined date can be excluded if the reliability of the data is poor or if significant changes are known to have occurred as a result of habitat modification. Similarly, records could be filtered to take into account particular biological attributes of the TI, such as host plant or season. Background Groups The background group for any particular TI is defined as those organisms likely to be targeted by the same collector or caught by the same collecting methods as the TI. Because records of background-group taxa are essentially used as a surrogate for searching or sampling effort for the TI, the choice of appropriate taxa for inclusion in the background group is important. They are not fixed and may be readily redefined to test the effects of adding or subtracting taxa or using filters to exclude inappropriate data (e.g., collections made during seasons in which the TI does not occur). The background group is not

Conservation Biology Volume 15, No. 3, June 2001

Ponder et al.

necessarily strictly a taxonomic group because it is intended to reflect the likely biases of collectors and collecting methods, and it may contain several taxonomic groups or parts thereof. For example, whereas some entomologists target mainly the group in which they have an interest (e.g., butterflies), others may collect all those insects that happen to be caught by the particular sampling methods employed (e.g., pitfall traps or foliage fogging). By contrast, land snails, which we used as an example, are generally collected as a group, either by eye (larger species) or from litter samples (smaller species). Because not all collectors employ both methods, we divided land snails into two independent background groups (based on size, not phylogenetic relationship) for analysis. Background Sampling Analysis Our basic premise is that if the background sampling effort is “adequate,” then the mapped distribution of a particular taxon may be presumed to accurately reflect its true distribution. Survey sites where background sampling has taken place but the TI is absent can be used as “pseudo-absences” (in lieu of true absence data) to define the distribution of the taxon. In contrast, if there is little or no background sampling effort for an area, it is unlikely that the distribution of the TI as described by the available records will be accurate. We used two approaches to assess the robustness of a TI’s distribution in terms of its background sampling effort, the first statistical and the second based on density surfaces. We illustrate these approaches with a dataset consisting of actual background data for land snails from northern New South Wales, overlain with hypothetical TI distributions based on commonly observed distributional patterns for narrow-range taxa. SPATIAL STATISTICS

A set of statistics was generated that described the spatial distribution of geographic points for each set of data (i.e., each TI and its background group). The program used for calculating these statistics is available from the second author. These statistics describe the spatial distribution of records and do not directly indicate the robustness of the background sampling data. They can be used to indicate problems with the data, however, such as excessive clustering, highly scattered distributions, or the presence of outliers (erroneously located collection sites)— problems that might require the data to be examined further or discarded. This can be particularly useful in instances where there are a large number of taxa to be screened, precluding individual visual examination. Within-group statistics describe the spatial distribution and relationships between points in each dataset.

Ponder et al.

They include the maximum, minimum, median, and mean separation (and standard deviations) of each point from every other point in the group, and each point from its nearest neighbor, as well as (1) the area occupied by the points; (2) the density of points; (3) the mean nearest-neighbor separation for a uniform distribution of the same number of points in the same-size area; (4) the ratio of the actual mean nearest-neighbor separation to that of a uniform distribution; and (5) the number of points separated by more than the mean nearestneighbor distance. These statistics are calculated for the TI, for the entire background group, and for those background-group records that fall within a certain radius of the TI records (a so-called “buffered” background group). The buffer— the size of which can be chosen by the user depending on the question at hand—is used because, in many analyses, the total background-group area is inappropriate to the question at hand. The buffer also serves to minimize the effect of variation in sampling intensity in the background group by standardizing the scale of analysis. This is necessary because sampling variation (clustering) increases with area, partly because sampling often occurs opportunistically and on a localized scale, and partly because distributions are usually not random. Between-group statistics were generated to describe the distribution of each TI’s points relative to those of its background group. These include the maximum, minimum, median, and mean separation of each TI point from every point in the background group, and of the nearest background-group neighbor of each TI point. Statistics were calculated with both the full backgroundgroup data and the buffered background data, as for the within-group statistics. Following analysis of these statistics, a TI may be excluded from further analysis, or tagged for checking if it contains outliers, if the data are considered excessively clustered, or if there appears to be inadequate background sampling based on the densities of points and distances between TI and background-group records. The thresholds for each of these parameters must be decided on a project or group basis, with consideration of the data sampling regime. DENSITY SURFACES

The next step involved the production of a contouredbackground sampling-density surface based on the background data for each TI. Nelson et al. (1990) used an analogous contouring method to show the density of angiosperm collections in the Amazon based on records per grid square. The density surface was created by removing the TI’s records from the background group and using the Arcview (Environmental Systems Research Institute, Redlands, California) “density” function to model a surface for which the values are projected numbers of

Evaluating Museum Data

651

collection records per square kilometer. These values are calculated by assigning a density to each grid-cell, taking into account other records within a user-defined distance (in our case 50 km). The density value is analogous to survey effort for the TI’s background group. The removal of the TI’s records prior to creation of the density surface is designed to generate a surface that is independent of sampling efforts for the TI. The rationale for this is that inclusion of the TI records results in misleading values for background sampling effort for some taxa, such as those that have been heavily collected at a particular location but have few background records in the vicinity. In the case of distinctly allopatric taxa, however, the TI records may need to be included because their removal will leave a “hole” in the sampling-density surface. A single figure reflecting the background sampling intensity for each TI can then be calculated by averaging the value of the density surface at each TI point. This calculation provides an objective value for background sampling that can be used to make modeling decisions. For instance, TI’s can be included or excluded from further analyses based on whether the background sampling density for their known range meets a user-defined minimum level.

Predictive Modeling Background data and the records for one or more TIs can be overlaid with various environmental parameters that may explain sampling gaps, such as extensive agricultural land, urban development, or a significant change in climate, geomorphology or, geology. Simple correlations such as these can be undertaken relatively easily by overlaying transparencies or using GIS technology for computer mapping. More sophisticated analysis requires computer modeling, as outlined in this section. Computer modeling enables testing of the observed distributions obtained from museum-collection data against predicted distributions modeled from abiotic (e.g., climatic and geological) data. Modeling has been used to predict the distributions of species (e.g., Nix 1986; Reid 1998; Regional Forest Agreement Steering Committee 1998), which can then be mapped by GIS. Species-distribution models can be designed for an optimum trade-off between accuracy (e.g., small grid cells) and available computational ability. The modeling algorithm we used is based on the BIOCLIM approach (Nix 1986), which uses climatic variables and GIS technology to map predicted distributions of TIs. VARIABLES

Climatic variables for species-distribution modeling are obtained from environmental surfaces generated by the

Conservation Biology Volume 15, No. 3, June 2001

652

Evaluating Museum Data

ESOCLIM program (Nix 1986; Busby 1991). Basic meteorological surfaces (e.g., temperature and precipitation) are used to derive a range of environmental surfaces— including variability and seasonality in rainfall, temperature, and radiation—that are more likely to reflect biological dependencies or critical limiting factors for the species. Other important variables, which can be extrapolated from a digital elevation model (DEM), include topographic indices such as ruggedness, slope, and aspect. Other useful variables include soil fertility, depth, and moisture, geology, and vegetation. These surfaces are used in turn to define environmental profiles for TIs based on available records. Many of these variables— particularly rainfall and temperature and their seasonality—have already been widely used to predict species distributions (e.g., Nix 1986; Austin & Meyers 1996; Ferrier & Watson 1997; Peters & Thackway 1998; Reid et al. 2001). MODELING DISTRIBUTIONS OF TAXA

We modeled predicted distributions of TIs from collection-site data and environmental variables using an Arcview script. The script read the value of each bioclimatic variable at each site, sorted and categorized them, and ranked the values for each variable in ascending order to create an environmental profile for the TI. Proportions of the profile (confidence intervals) were used to exclude outlying values, depending on the level of confidence in the spatial accuracy of sites. Outliers may give a misleading environmental profile and affect the geographical distribution predicted by the model. Commonly, the 5–95% confidence interval is used to eliminate the influence of outliers (e.g., Nix 1986; Reid et al. 2001), although this is an arbitrary setting. These intervals were applied to the environmental surfaces to create a map of the modeled or predicted TI distribution. Distributions were further refined by adding other GIS layers, such as geology and vegetation. For instance, a TI may be associated with specific substrate or vegetation types which, if known, allows elimination of predicted distributions that fall outside these types. Predicted distributions should be verified by appropriate experts and, where feasible, by field collecting.

Ponder et al.

& Watson 1997) delivers a measurement of the robustness of both the model and the raw data, ensuring that the model can provide consistent results given changes in the site information. Bootstrapping involves quantification of the response of the model to subsets of the original data, the measurement being the percent overlap between models at successive levels of subsampling (e.g., 100%, 95%, 90%, etc.), averaged over a number of random samples at each level. Low levels of agreement between the subset models indicate the presence of outliers or of poor sampling coverage.

Results The distributions of the five hypothetical narrow-range land snails—TIs a, b, c, d, and e—were analyzed based on the same background group (actual land-snail data). The relationship among these species and the background data reflect patterns commonly found in nature (Fig. 1). Background Sampling The TIs a and d appear to have better levels of background sampling than either b or c, even though c has more records than d (Fig.1). The TI e occurs in both well-sampled and poorly sampled areas. Analysis of this example, through calculation of spatial statistics and generation of a background sampling-density surface, is described below.

ROBUSTNESS OF MODELS

We assessed the robustness of the TI’s distribution model by evaluating the model either on an iterative basis or by bootstrapping. In iterative assessment, potential outliers are examined after each run, allowing their identification and location to be checked against the original collection data. Modeling can indicate incorrectly identified or databased records, which makes it extremely useful for museum-collection data. Bootstrapping (sampling without replacement; Ferrier

Conservation Biology Volume 15, No. 3, June 2001

Figure 1. Distributions of five hypothetical taxa of interest (TI, a–e), with the background records (crosses) based on actual data for land snails in northern New South Wales.

Ponder et al.

Table 1.

Evaluating Museum Data

Sample of within-group statistics describing the spatial distribution and relationships between points in each data set.*

TI

Data set

No. records

a

TI 50-km buffer BG TI 50-km buffer BG TI 50-m buffer BG TI 50-km buffer BG TI 50-km buffer BG

8 89 317 3 34 322 6 68 319 3 64 322 3 148 322

b c d e

653

MeanWithinGp (km)

MaxWithinGp (km)

MeanNearest (km)

MaxNearest (km)

Area (km2)

Density (km2)

UniformNearest (km)

Actual/ Uniform (%)

No.  MeanNearest

5.6811 47.1323 99.6857 2.8112 50.8413 99.1236 6.0263 50.0836 99.1385 21.392 42.1471 98.5816 63.2289 75.8079 99.0628

12.2536 119.7919 235.9132 3.7971 99.3243 235.9132 13.7797 109.0267 235.9132 30.9614 103.4356 235.9132 80.6653 176.5659 235.9132

2.3796 0.1525 0.1048 1.8759 1.2992 0.1047 2.0977 0.2572 0.1054 15.0772 0.2604 0.1053 53.5686 0.1093 0.1047

4.4577 0.2927 1.7805 2.7100 4.5999 1.7805 8.8457 0.9089 1.7805 18.1605 2.6361 1.7805 57.3369 0.2927 1.7805

42 8634 34824 3 7502 34824 32 8510 34824 109 7392 34824 1724 19275 34824

0.1905 0.0103 0.0091 1.000 0.0045 0.0092 0.1875 0.008 0.0092 0.0275 0.0087 0.0092 0.0017 0.0077 0.0092

2.2913 9.8494 10.4812 1.000 14.8542 10.3995 2.3094 11.1869 10.4483 6.0277 10.7471 10.3995 23.9722 11.4121 10.3995

103.8521 1.5481 1.0002 187.5905 8.7461 1.0065 90.832 2.2988 1.0087 250.1314 2.4234 1.0124 223.4613 0.9577 1.0065

3 34 3 1 5 3 1 10 3 1 7 3 1 7 3

*Statistics calculated for each taxon of interest (TI, a–e), its “buffered” background group (BG, all points within 50 km of the TI records), and full BG. MeanWithinGp and MaxWithinGp, average and maximum distances between points in the group; MeanNearest and MaxNearest, average and maximum distances to the nearest neighbor of each point; UniformNearest, nearest-neighbor separation that would occur if the points were distributed uniformly; Actual/Uniform, ratio of the actual nearest-neighbor separation (MeanNearest) to that of a uniform distribution (UniformNearest); No.  MeanNearest, number of points separated by more than the mean nearest neighbor distance.

SPATIAL STATISTICS

Subsets of the within-group and between-group statistics were calculated for each TI and its background group (Tables 1 & 2 respectively). Within-group statistics can indicate the presence of outliers or clustering in the distribution of a TI or its background group. A small number of records (50%) with more than the average nearest-neighbor separation (no.  MeanNearest [average distance to nearest neighbor of each point]) suggests the presence of one or more outliers. Close to half (3/8) of the records of TI a have a greater than average nearest-neighbor separation, suggesting a fairly uniform distribution, whereas only one-sixth of the records of c do so, indicating an outlier (Table 1). For TIs with few records (e.g., b, d, and e), comparison of

MaxNearest (maximum distance to nearest neighbor of each point) to MeanNearest may help flag outliers. Clustering in a TI may indicate either outliers or a disjunct distribution. The degree of uniformity or clustering of records may be shown by the ratio of the actual nearest-neighbor separation to that of a uniform distribution (Actual/Uniform statistic), with values of around 100 generally indicating a more uniform distribution and lesser values indicating greater clustering. The high values for TIs b, d, and e are artefacts (each has only three records), but TI c has a value of 90.8, indicating some clustering. The value for TI a (103.9), however, suggests that these points are distributed more uniformly. For a TI, high density and low separation of points (MeanNearest or the average [MeanWithinGp] and maxi-

Table 2. Sample of between-group statistics describing the distribution of each taxon of interest’s (TI) points relative to those of its background group (BG).* TI

Background data

TI records

BG records

MeanBtwnGp

SD(MeanBtwnGp)

MinBtwnGp

a

50-km buffer BG 50-km buffer BG 50-km buffer BG 50-km buffer BG 50-km buffer BG

8 8 3 3 6 6 3 3 3 3

89 317 34 322 68 319 64 322 148 322

34.8486 80.5213 37.9954 77.0229 42.107 88.153 33.4761 105.961 64.7971 80.0879

19.8957 40.104 10.1238 21.592 9.8964 34.0319 18.4945 48.7315 31.0023 32.3276

0.912 0.912 7.4944 7.4944 14.594 14.594 1.6987 1.6987 1.905 1.905

b c d e

*Comparison of each TI (a–e) with its buffered and full BG: MeanBtwnGp, average distance between TI and BG points; SD(MeanBtwnGp), standard deviation of this; MinBtwnGp, minimum distance between a BG and a TI point.

Conservation Biology Volume 15, No. 3, June 2001

654

Evaluating Museum Data

mum [MaxWithinGp] distances between points in the group) indicate a well-sampled taxon, whereas low density and high separation suggest a more widely distributed “rare” taxon. The TI e falls into the latter category, with density-to-separation ratios an order of magnitude higher than those of the other TIs. Thus, it is probable that e has not been sampled adequately over its range and that the available data do not accurately reflect its true distribution. The density of buffered background points is highest for TI a (0.0103/km2) and lowest for b (0.0045/km2) (Table 1). Higher confidence in background sampling is given by a high density and low separation (MeanWithinGp) of buffered background points. The buffered backgrounds for TIs c, d, and a were better than those of e and, particularly, b. Based on the between-group statistics TIs a and d had the lowest values for average distance (MeanBtwnGp) and minimum distance (MinBtwnGp) between TI and the background group, indicating that the background records were closer to the records for these TIs (Table 2). The TI c and, to a lesser extent, b have higher values, suggesting insufficient sampling in the immediate vicinity and low confidence in the accuracy of their distributions. The TI e has a high standard deviation, indicating variability in the distances between its own and background-group records. A comparison of the separation between TI and background-group points (e.g., MeanBtwnGp) with the density of buffered background-group points (Fig. 2) provides the best indication of confidence in the adequacy of background sampling, and hence robustness of a TI’s mapped distribution. The TIs a and d have the highest densities of buffered background points and the lowest mean separation of these points from their own records, reflecting the higher sampling intensities in the areas surrounding these TIs. The TIs a and d have the most reliable levels of background sampling (Fig. 2). The TI c has a possible outlier (no.  MeanNearest  1/6), but the distances between its records and those of its background group (Fig. 2) indicate a better level of sampling than that of TI b, which has the poorest background sampling of any taxon. The level of sampling around TI e is highly variable and is probably inadequate to accurately determine its distribution. DENSITY SURFACES

An example of the density of background sampling intensity can be shown with TI a removed (TIs b–e are included as part of the background) (Fig. 3). The single-figure values for background sampling intensity calculated for each TI were a  0.027586, b  0.001623, c  0.000477, d  0.012117, and e  0.010391. These values were calculated by averaging the value of the TI’s density surface (Fig.2) at each TI point.

Conservation Biology Volume 15, No. 3, June 2001

Ponder et al.

Figure 2. A plot of the mean distances between taxon of interest (TI) and buffered background-group (BG) points against the density of sampling (i.e., number of data points per km2) for the buffered BG.

The TI a has the highest mean value for background sampling (0.0276); TI d (0.0121) had the second-highest. Both of these are two orders of magnitude or more higher than either b (0.0016) or c (0.0005), even though c has the second-greatest number of records. Thus, the levels of background sampling for b and c appear to be too low to substantiate the distribution given by the available points. The intermediate value obtained for e (0.0103) reflects the fact that its records occur in both well- and poorly sampled areas and emphasizes the need to consider variation in sampling across a TI’s distribution. These figures support the conclusion gained from visual inspection of the data, but supply a quantitative measure of background sampling intensity.

Predictive Modeling After checking and (if appropriate) removing outliers, testing of the 100% model at different bootstrapping levels has shown that repeated random samples of the records can lead to high agreement in subset models even with samples as small as 50% of the original points. Using the methods described, for a well-sampled endangered species of land snail (Meridolum corneovirens; Pfeiffer 1851) in Western Sydney (for distribution map; see Little 1999), we found 98% agreement in the model at the 50% level, and 81% with only 25% of the original records. This indicates the robustness of the modeled distribution, even when only one-quarter of the available records are used.

Ponder et al.

Figure 3. Contour surface of background sampling for taxon of interest (TI) a. Circle shows the 50-km radius (buffered background) around TI a.

Discussion Museum and herbarium data have played an important role in the biodiversity assessment and conservation of many vertebrate groups and vascular plants, and it is emphasized in international programs such as the Global Taxonomic Initiative (American Museum of Natural History 1999). For invertebrates, information resides mainly in museum collections. More effective utilization of these data will be enhanced by development of more rigorous approaches. Our methods differ from those of previous studies, which have focused mainly on the mapping of grid-based data, because we have assessed the robustness of distributions of taxa based on point data from museum collections. Somewhat analogous methods for describing patterns in distribution density (Rossi et al. 1992), originally developed for mining, have been adapted for ecological purposes from geostatistics (Royle et al. 1980). They differ in that they rely on standardized sample spacing (lag distance) and thus on grid-based surveys in which sampling effort and spatial arrangement are regular. Our methods are not dependent on area and can be tailored for use at different scales and in different environments, depending on data availability. The use of background sampling analysis, combined with predictive modeling, provides a means of independently testing the robustness of the known distribution of a particular taxon. Both approaches identify areas that can be field-tested: the first indicates gaps in the sampling and the second indicates areas with similar environmental profiles where the TI might be expected to occur. Field testing of these latter areas should be prioritized accord-

Evaluating Museum Data

655

ing to low or absent background sampling or the presence of suitable or original habitat (e.g., forest cover). Other potential applications of museum-collection data for which our methods could be used include comparison of current with past distributions, assessment of distributions relative to current or proposed land-use options, and assessment of the relationship of distributions to various environmental variables. In addition, the application of environmental models to museum distributional data could be extended to analyze potential changes in species distributions caused by changes in temperature, rainfall, or vegetation. Limitations of these methods include the fact that there is reduced confidence in predictive power with small numbers of records and when the data are highly clustered. Another is that the background-sampling method will not be able to verify the distributions of isolated allopatric taxa surrounded by large areas of habitat unsuitable for members of their background group(s). This problem can be partially overcome through use of appropriate environmental overlays, in an a posteriori fashion, that exclude the unsuitable areas. The fundamental premise of the background-sampling analysis (that if the intensity of background sampling is “adequate,” then the mapped distribution probably reflects the true distribution) involves some fairly major biological assumptions—that seasonality, habitat fidelity, and annual population variation are understood. These assumptions cannot always be made, especially for insects and other taxa with high seasonality and host specificity. This problem can be partly overcome through application of appropriate filters to test particular biological assumptions, such as seasonality and host-plant preference. In the same way, data can also be filtered to include or exclude particular sampling methods. Other limitations of these methods include the need for (1) accurate species-level identifications for at least the TI (accurate identification of other taxa in the background group is not essential); (2) accuracy of the specimen-locality data; (3) appropriate choice of taxa for background groups, requiring a reasonable knowledge of collectors, collection methods, and the biology of component groups; and (4) availability of environmental layers for the modeling assessment. These taxonomic and data requirements may lead to some difficulties in areas for which detailed environmental data, or the appropriate taxonomic expertise, are unavailable. Nevertheless, even if the outcomes may be less reliable, the methodology can still indicate sampling gaps and give indications of the reliability of distributions by means of the statistical procedures and/or density surfaces. It is important that methodologies for evaluating museum data continue to be developed and enhanced but, equally, that their value be more widely recognized and that they be made more accessible through databasing

Conservation Biology Volume 15, No. 3, June 2001

656

Evaluating Museum Data

and improved data quality. The move toward online access to museum-collection data should eventually result in a global biodiversity facility of immense importance. The provision of adequate resources is essential, and taxonomic expertise and ongoing field work will be indispensable to improving and expanding these data.

Acknowledgments We thank J. Kelly for developing the program to provide the statistics. Many people, in particular the collection managers at the Australian Museum, have contributed to the Narrow Range Endemics Program within which the methods outlined in this paper were developed. D. Faith commented on an early draft of the manuscript, and C. Reid made useful comments on a semifinal draft. We also thank L. Wilkie for statistical comment.

Literature Cited American Museum of Natural History. 1999. The Global Taxonomic Initiative: using systematic inventories to meet country and regional needs. Centre for Biodiversity and Conservation, American Museum of Natural History, New York. Austin, M. P., and J. A. Meyers. 1996. Current approaches to modelling the environmental niche of eucalypts: implication for management of forest biodiversity. Forest Ecology and Management 85:95–106. Bickel, D. J. 1999. What museum collections reveal about species accumulation, richness, and rarity: an example from the Diptera. Pages 174–181 in W. Ponder and D. Lunney, editors. The other 99%: the conservation and biodiversity of invertebrates. Royal Zoological Society of New South Wales, Mosman. Busby, J. R. 1991. BIOCLIM: a bioclimate analysis and prediction system. Pages 64–68 in C. R. Margules and M. P. Austen, editors. Nature conservation: cost effective biological surveys and data analysis. Commonwealth Scientific, Industrial, and Research Organization, Canberra, Australia. Colwell, R. K., and J. A. Coddington. 1994. Estimating terrestrial biodiversity through extrapolation. Philosophical Transactions (Royal Society of London) B 345:101–118. Cotterill, F. P. D. 1995. Systematics, biological knowledge and environmental conservation. Biodiversity and Conservation 4:183–205. Cotterill, F. P. D. 1997. The second Alexandrian tragedy, and the fundamental relationship between biological collections and scientific knowledge. Pages 227–241 in J. R. Nudds and C. W. Pettitt, editors. The value and valuation of natural science collections. Proceedings of the International Conference, Manchester, 1995. The Geological Society, London. Davis, P. 1996. Museums and the natural environment: the role of natural history museums in biological conservation. Leicester University Press, London. Disney, R. H. L. 1986. Assessments using invertebrates: posing the problem. Pages 271–293 in M. B. Usher, editor. Wildlife conservation evaluation. Chapman and Hall, London. Drinkrow, D. R., and M. I. Cherry. 1995. Anuran distribution, diversity and conservation in South Africa, Lesotho and Swaziland. South African Journal of Zoology 30:82–90. Fagan, W. F., and P. M. Kareiva. 1997. Using compiled species lists to make biodiversity comparisons among regions: a test case using Oregon butterflies. Biological Conservation 80:249–259.

Conservation Biology Volume 15, No. 3, June 2001

Ponder et al.

Ferrier, S., and G. Watson. 1997. An evaluation of the effectiveness of environmental surrogates and modelling techniques in predicting the distribution of biological diversity. Consultancy report prepared by New South Wales National Parks and Wildlife Service. Department of Environment, Sport & Territories. Environment Australia, Canberra. Hawksworth, D. L., and L. A. Mound. 1991. Biodiversity databases: the crucial significance of collections. Pages 17–29 in D. L. Hawksworth, editor. The biodiversity of microorganisms and invertebrates: its role in sustainable agriculture. Proceedings of the first workshop on the ecological foundations of sustainable agriculture 1991. Commonwealth Agricultural Bureaux International, Oxon, United Kingdom. Heywood, V. H., editor. 1995. Global biodiversity assessment. Cambridge University Press, Cambridge, United Kingdom. Hoagland, K. E. 1989. Socially responsible: in the 1990s natural history museums will focus their efforts on maintaining the diverse life of a healthy planet. Museum News (September/October): 50–52. Kremen, C., R. K. Colwell, T. L. Erwin, D. D. Murphy, R. F. Noss, and M. A. Sanjayan. 1993. Terrestrial arthropod assemblages: their use in conservation planning. Conservation Biology 7:796–808. Kress, W. J., W. R. Heyer, P. Acevedo, J. Coddington, D. Cole, T. L. Erwin, B. J. Meggers, M. Pogue, R. W. Thorington, R. P. Vari, M. J. Weitzman, and S. H. Weitzman. 1998. Amazonian biodiversity: assessing conservation priorities with taxonomic data. Biodiversity and Conservation 7:1577–1587. Little, S. J. 1999. Implications for threatened invertebrates in the New South Wales planning system. Pages 361–368 in W. Ponder and D. Lunney, editors. The other 99%: the conservation and biodiversity of invertebrates. Royal Zoological Society of New South Wales, Mosman. Majer, J. D., editor. 1987. The role of invertebrates in conservation and biological survey. Western Australian Department of Conservation and Land Management, Perth. McCarthy, M. A. 1998. Identifying declining and threatened species with museum data. Biological Conservation 83:9–17. Myers, N. 1988. Threatened biotas: ‘hot spots’ in tropical forests. Environmentalist 8:187–208. Myers, N. 1990. The biodiversity challenge: expanded hot-spots analysis. Environmentalist 10:243–256. Nelson, B. W., C. A. C. Ferreira, M. F. da Silva, and M. L. Kawasaki. 1990. Endemism centres, refugia and botanical collection density in Brazilian Amazonia. Nature 345:714–716. New, T. R. 1998. Invertebrate surveys for conservation. Oxford University Press, New York. Nielsen, E. S., and J. G. West. 1994. Biodiversity research and biological collections: transfer of information. Pages 101–121 in P. L. Forey, C. J. Humphries, and R. I. Vane-Wright, editors. Systematics and conservation evaluation. Systematics Association special volume 50. Clarendon Press, Oxford, United Kingdom. Nix, H. 1986. A biogeographic analysis of Australian elapid snakes. Pages 4–15 in R. Longmore, editor. Atlas of elapid snakes of Australia. Australian flora and fauna Series 7. Bureau of Flora and Fauna, Canberra, Australia. Peters, D., and R. Thackway. 1998. A new biogeographic regionalisation for Tasmania. Report prepared for the National Reserve System Program Component of the Natural Heritage Trust. Tasmanian Parks and Wildlife Service, Geographic Information System Section, Hobart, Tasmania. Ponder, W. 1999. Using museum collection data to assist in biodiversity assessment. Pages 253–256 in W. Ponder and D. Lunney, editors. The other 99%: the conservation and biodiversity of invertebrates. Royal Zoological Society of New South Wales, Mosman. Ponder, W., and D. Lunney (editors). 1999. The other 99%: the conservation and biodiversity of invertebrates. Transactions of the Royal Zoological Society of New South Wales, Mosman.

Ponder et al.

Prendergast, J. R., R. M. Quinn, J. H. Lawton, B. C. Eversham, and D. W. Gibbons. 1993a. Rare species: the coincidence of diversity hotspots and conservation strategies. Nature 365:335–337. Prendergast, J. R., S. N. Wood, J. H. Lawton, and B. C. Eversham. 1993b. Correcting for variation in recording effort in analyses of diversity hotspots. Biodiversity Letters 1:39–53. Reid, W. V. 1998. Biodiversity hotspots. Trends in Ecology and Evolution 13:275–280. Reid, C., P. Cranston, and I. Reid. 2001. Biogeographic and historical determinants of dung beetle richness in tropical rainforests of northeast Queensland. In press in C. Moritz, editor. Rainforest: past, present, future. Chicago University Press, Chicago. Regional Forest Agreement Steering Committee. 1998. Comprehensive regional assessment: a regional forest agreement for Western Australia. Volume 1. Commonwealth of Australia and Western Australian Governments, Perth. Rossi, R. E., D. J. Mulla, A. G. Journel, and E. H. Franz. 1992. Geostatistical tools for modelling and interpreting ecological spatial dependence. Ecological Monographs 62:277–314.

Evaluating Museum Data

657

Royle, A. G., I. Clark, P. I. Brooker, H. Parker, A. G. Journel, J. M. Rendu, R. I. Sandefur, D. C. Grant, and P. Mousset-Jones. 1980. Geostatistics. McGraw Hill, New York. Skelton, P. H., J. A. Cambray, A. Lombard, and G. A. Benn. 1995. Patterns of distribution and conservation status of freshwater fishes in South Africa. South African Journal of Zoology 30:71–81. Väisänen, R., and K. Heliövaara. 1994. Hot-spots of insect diversity in northern Europe. Annales Zoologici Fennici 31:71–81. Vane-Wright, R. I., C. R. Smith, and I. J. Kitching. 1994. Systematic assessment of taxic diversity by summation. Pages 309–326 in P. L. Forey, C. J. Humphries, and R. I. Vane-Wright, editors. Systematics and conservation evaluation. Systematics association special volume. Oxford University Press, Oxford, United Kingdom. Williams, P. H., G. T. Prance, C. J. Humphries, and K. S. Edwards. 1996. Promise and problems in applying quantitative complementary areas for representing the diversity of some Neotropical plants (families Dichapetalaceae, Lecythidaceae, Caryocaraceae, Chrysobalanaceae and Proteaceae). Biological Journal of the Linnean Society 58:125–157.

Conservation Biology Volume 15, No. 3, June 2001

Suggest Documents