Efficient floristic inventory for the assessment of tropical tree diversity: A comparative test of four alternative approaches

Forest Ecology and Management 237 (2006) 564–573 www.elsevier.com/locate/foreco

Efficient floristic inventory for the assessment of tropical tree diversity: A comparative test of four alternative approaches James E. Gordon a,*, Adrian C. Newton b a

Center for Tropical Plant Conservation, Fairchild Tropical Botanic Garden, 11935 Old Cutler Road, Coral Gables, FL 33156-4242, USA b School of Conservation Sciences, Bournemouth University, Talbot Campus, Poole, Dorset BH12 5BB, UK Received 29 June 2006; received in revised form 3 October 2006; accepted 4 October 2006

Abstract Biodiversity inventory as a tool for guiding conservation planning at a local scale is under used, especially in tropical countries where technical capacity is often limited. Here four inventory protocols (one ad hoc, one fixed count and two fixed area methods) are tested for their efficiency and statistical robustness when applied to the woody floras of a Mexican seasonally dry tropical forest. They are tested for their ability to distinguish between sites not only by number of species (richness), but also by number of threatened species, and finally in their application to complementary reserve selection exercises. Whilst the ad hoc method is shown to have the advantage of high efficiency and is technically undemanding, where possible the statistically more robust fixed count or fixed area methodologies are recommended. However, the much used fixed area method of Gentry [Gentry, A.H., 1982. Patterns of neotropical plant species diversity. Evol. Biol. 15, 1–84] that uses 2 m  50 m plots is shown to be inefficient. Only when comparison with other previously published assessments is a priority should the 2 m  50 m fixed area method be considered, and then efficiency may be increased by widening the plots and distinguishing between those stems that fall within a notional 2 m plot and those outside of it. Both prioritisation by number of species and complementarity analysis, using the ‘greedy’ algorithm, are shown to be sensitive to choice of inventory protocol, suggesting that consideration is given to how species lists are obtained in reserve selection exercises. # 2006 Elsevier B.V. All rights reserved. Keywords: Biodiversity assessment; Rapid botanical assessment; Reserve selection; Tropical dry forest; Selection algorithm

1. Introduction There is general agreement that biodiversity conservation should be guided by biodiversity assessment (Margules and Pressey, 2000; Phillips et al., 2003; Pressey et al., 1993; Royal Society, 2003) and this is recognised by international policy processes such as the Convention on Biological Diversity. Biodiversity assessment is, however, a broad term that has been used by different authors to include a number of different activities (e.g. Cantu´ et al., 2004; Dudley and Jeanrenaud, 1998; Hilton-Taylor et al., 2000; Lawrence, 2002; UNEP-WCMC, 2003). A distinction is made here between biodiversity inventory and biodiversity monitoring, both of which may be considered part of biodiversity assessment. Inventory, a static process and the subject of this article, is used to identify ‘the

* Corresponding author. Tel.: +1 305 667 1651; fax: +1 305 588 0383. E-mail addresses: [email protected] (J.E. Gordon), [email protected] (A.C. Newton). 0378-1127/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.foreco.2006.10.002

specific conservation target that the project ultimately would like to influence’ (Salafsky et al., 2002) whilst monitoring is a dynamic process that aims to detect changes in the status of biodiversity over time. At global and regional scales considerable progress has been made in assessing conservation priorities for broad habitat types (e.g. Dinerstein et al., 1995) and for specific taxa or life forms (e.g. Ceballos et al., 1998; Davis et al., 1997; Heywood, 1995; Myers et al., 2000). However, at local scales, within habitats regarded as conservation priorities, there is often little use of biodiversity inventory with the result that decisions about what to protect are driven primarily by socio-economic criteria (Gordon, 2006; Gordon et al., 2006). Such selection results in reserve networks that are considered inefficient as they inevitably contain fewer species than the theoretical maximum (Pressey, 1994). Hence there remains a divide between theoretical best practice and site selection as it usually occurs (Prendergast et al., 1999). This is especially so in lower-income countries where a disproportionately high number of the world’s most diverse terrestrial ecosystems are found (Myers

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573

et al., 2000) but where the resources and technical capacity needed for biodiversity inventory are severely constrained. This places a priority on identifying efficient and relatively simple methods of floristic inventory. Various field methodologies for tree diversity inventory have been proposed. Fixed area methods are the most widely used, with the commonest rapid assessment technique perhaps being the 0.1 ha method popularised by Gentry (1982) that uses 10 repetitions of 2 m  50 m ‘strip transects’. Phillips et al. (2002) compiled data from over 200 vegetation samples based on variants of this protocol from around the world. In one of the few attempts to directly measure and compare the efficiency of inventory methodologies, Phillips et al. (2003) tested the Gentry method against the use of 1 ha plots and showed that the former was more efficient as measured by the number of species encountered against time searching, the 1 ha plot being impractically large for rapid floristic assessment. Furthermore, the 2 m  50 m method was shown to be more statistically powerful because more repetitions per unit time were achieved. They did not, however, test this method against other floristic sampling methods specifically proposed as rapid methodologies. More recently Gordon and Newton (2006) have argued that the size of samples that result from 10 repetitions of 2 m  50 m plots do not capture sufficient within site variation to allow confident comparison between forests located across broad geographical areas. Hall (1991) made a case for methods based on a fixed tree counts in which sampling effort is controlled by selecting a fixed number of trees per point sample (‘plot’) with no requirement for plot demarcation. He used such a method to effectively survey montane forest in Africa, but provided no direct comparison of its efficiency with other methods. Condit et al. (1998) supported fixed tree count methods on statistical grounds. They argued that by comparing equal numbers of stems the resulting diversity indices were not prone to biases resulting from differences in density that affect fixed area methods. Stern (1998) tested 2 m  50 m plots against fixed tree counts, which she referred to as variable area transects, within completely inventoried 1 ha plots in Peruvian Amazonia. She concluded that the fixed count plots were more flexible, particularly when different vegetation structures were encountered, but that strip transects had the advantage of being comparable to assessments from many other sites worldwide. However, she went on to question whether estimating species diversity is in itself worth the considerable investment required and suggested that a simple checklist of important or uncommon species, along with a brief structural description, might be a more useful conservation tool. Variations on this checklisting or ad hoc approach are less well tested but are widely used, for example in Conservation International’s Rapid Biodiversity Assessments (e.g. Schulenberg et al., 1999). With ad hoc methods, the control of sampling effort, which in other methodologies is done by fixing plot area or the number of trees per point sample is minimal. Because of this lack of control, comparisons between the resulting checklists are difficult, and the lack of repeated sampling at each site prevents quantification of the variability associated

565

with the derived estimates, that is, confidence limits cannot be calculated. Furthermore, checklisting approaches do not easily allow accurate estimation of abundances of the species encountered. However their advantages, principally of economy, have been highlighted by Droege et al. (1998) with respect to the monitoring of plants and animals, and they have been applied to the assessment of tree diversity by Gordon et al. (2004) and Hawthorne and Abu-Juam (1995). The field methodologies under consideration here vary in the way in which sampling effort is controlled. However, gains in efficiency can also be made by altering the sampling frame. When a restricted sampling frame is used in biodiversity assessment with the expectation that it will represent a wider subset of total biodiversity, the former is often called a surrogate of the latter. Such surrogates include: indicator species and higher taxa (Balmford et al., 1996; Chase et al., 2000; Kerr et al., 2000), functional groups (Vanclay et al., 1997) and disturbance regime (Sagar et al., 2003). A comparison of some of these surrogates is provided by Higgins and Ruokolainen (2004). The comparison of sites by species richness (e.g. Kerr et al., 2000) is a relatively crude basis for prioritisation of areas for conservation, whereas limiting comparisons to species considered to be threatened offers the possibility of focusing conservation resources on those species that most need it. Here consideration is also given to the performance of inventory protocols when only the subset of those species that are threatened, here defined by having restricted natural ranges, are used. The concept of complementary reserve selection is also now well established, at least in the literature (Gaston and Rodrigues, 2003; Margules et al., 1988; Pressey et al., 1993) as a means of ensuring reserve networks contain all species locally recorded. In complementarity analyses, a site’s importance is not measured by species richness but by the number of species that it contains that are not already represented elsewhere in the network. Whilst much attention has been paid to the development of increasingly sophisticated algorithms for use in complementarity analysis (e.g. Briers, 2001; Rodrigues and Gaston, 2002) less emphasis has been given to the sensitivity of their outputs to variation in input data, as provided by different inventory approaches. Here, four different inventory protocols for the characterisation of dry forest tree diversity are compared for their efficiency in guiding conservation planning. The aim is to compare fixed area, fixed count and ad hoc methods for the rapid inventory of tropical forest tree and shrub diversity in eight seasonally dry tropical forests sites in southern Mexico. Efficiency, given limited resources, is fundamental to the comparison of the protocols used, so the effort required by each method (person-minutes) compared to the number of species or restricted range species captured is key to the analysis presented, however consideration is also given to the statistical robustness of the data captured. In contrast to Gordon and Newton (2006), the emphasis here is on comparison between forests within a locality. The underlying assumption is that the locality has been chosen for conservation investment, and the objective of inventory is the identification of priority sites

566

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573

within it. Thus, this type of assessment has to distinguish between neighbouring forests that might be expected to be quite similar in composition. The distinctions that are made are based on comparison of species richness and number of restricted range species and by complementarity analysis. 2. Methodology

circumscribe the entire perimeter of the plot. The 2 m width is, however, arbitrary and the same advantage can be derived from wider plots. Hence the second of the four protocols was also a fixed area method in which the effect of altering plot size was tested by increasing the width of each to 6 m (3 m either side of the line) making a total area sampled of 0.45 ha per site. The 6 m width was chosen to maintain the advantages of narrow strip plots whilst greatly increasing the sample unit size.

2.1. Study site Eight areas of seasonally dry tropical forest were inventoried in 2002 in the municipality of Santa Marı´a Huatulco in Oaxaca, Southern Mexico. The tree and shrub diversity in this area is similar at familial and generic levels to other Mesoamerican dry forests and is considered in detail by Salas-Morales et al. (2003) and Gordon et al. (2004). From the mosaic of forest patches in the locality of interest eight sites were selected randomly with the precondition that each should contain closed canopy forest. Some of the sites were mid-succession secondary forest with the ages of the youngest estimated to be between 15 and 20 years. Other more mature sites may have been primary, although none could confidently be said to be unaffected by anthropogenic disturbance as the region has a history of commercial logging (Gordon et al., 2005) and extensive use by local farmers. The forest patches sampled varied considerably in size from under 50 ha to over 300 ha and their altitudes ranged from 20 m to 200 m. a.s.l. Sampling within the patches was carried out at least 100 m from the forest edge and riverine forest was excluded. One forest, site 8, was determined by visual inspection and prior to surveying to be the most floristically distinct of the sites, being atypically semideciduous and dominated by two species, Astronium graveolens Jacq. (Anacardiaceae) and Guarea excelsa Kunth (Meliaceae) that were absent from, or very rare in, the other sites. This forest patch was also taller, with a canopy height estimated to be around 15 m, the other seven sites varying between 10 m and 12 m. 2.2. Protocols 2.2.1. Fixed area The first fixed-area protocol adopted is a variant of the methodology popularised by Gentry (1982) in which 10 repetitions of 2 m  50 m strip plots are used but from which lianas were excluded and the lower diameter limit for inclusion of trees and shrubs was set at 5 cm, as it was the other protocols. This limit was set in order to capture the woody diversity of the canopy and sub-canopy but it is essentially arbitrary. Plots were randomly placed and, for comparative purposes, the number of plots was increased to 15 making a total area sampled of 0.15 ha per site. This methodology is henceforth called the 2 m  50 m protocol. The practical advantage of narrow plots is that they require only a line of 50 m to be cleared in the forest from which it can be easily determined, either by visual inspection or by rapid measurement, which stems fall within 1 m measured perpendicular to either side of the line. Thus, there is no need to

2.2.2. Fixed count Fixed count plots, here synonymous with point samples or variable area plots, require no plot demarcation and therefore are potentially more efficient than fixed area methods. Following the recommendations of Condit et al. (1998) that fixed count methods should have a minimum of 100 stems per sample, and of Hall (1991) that 15 trees per plot should be used, a protocol was devised in which each plot comprised the 15 trees closest to a central point. Fifteen randomly placed repetitions of this plot type were used at each site making a total sample size of 225 trees. The methodology is here called the fixed count protocol. 2.2.3. Ad hoc The ad hoc protocol was kept deliberately informal, it amounting to no more than a checklisting exercise in which the assessment team surveyed the area of forest sampled, starting at its perimeter and circling inwards until the team decided, subjectively, that no more new species were likely to be found. To assist with the comparative analysis the time at which each species was encountered was recorded. 2.2.4. General The inventory of a site started with the ad hoc protocol, the execution of which determined the area in which the randomly placed plots of the other methodologies would be placed. This ensured that each protocol sampled the same area of forest at each site. All operations were timed to facilitate efficiency comparisons. The field team was comprised of a data recorder, a tree spotter and a labourer. All three members of the team had prior knowledge of the local tree flora and therefore contributed to species identification. Collection of vouchers, when necessary, rarely required tree climbing because of the low stature of the forests and this therefore had an insignificant effect on efficiency. The vast majority of the species encountered had not been subjected to IUCN red listing, so a practical and quantitative assessment of threat status was used. The extinction risk faced by these species was approximated by weighting each in inverse proportion to its range size. This may appear to be a simplification of threat; abundance per unit area, rate of habitat loss and other factors are often considered when evaluating the threat status of a species, however for most of the species dealt with such information has never been collected. Range size estimations were based on herbarium specimen information held in Mexico’s National Herbarium, MEXU, and in the Tropicos database of the Missouri Botanic Gardens. Whilst biases exist in herbarium collections, in working with the

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573

largest regional herbarium, MEXU, and the database that manages data for the Flora Mesoamericana project we consider that a reasonable estimate of the current known range of each species was obtained. Given the variable precision with which herbarium specimen localities are recorded, the number of Mexican states/Central American countries in which a species had been collected was used to estimate distributions. The first step in creating a weighting was to calculate a relative area for each state and country by dividing the land area of each of those political entities by the land area of Oaxaca, thus giving Oaxaca a relative area of 1. To create a score for each species, the total relative land area of the states/countries in which each species was found was calculated and inverted. The highest score obtained for any species was 1 (for Oaxacan endemics) with all other species, which must be known from Oaxaca and at least one other state/country, scoring greater than 0 and less than 1. In this way, the species of most restricted range were given the highest weighting. Using a similar method, Gordon et al. (2004) classified approximately 20% of the Mesoamerican dry forest species they identified in this region as being of conservation concern and therefore the most restricted 20% of species were selected for the analysis of each protocol’s performance in capturing threatened species. 3. Analysis For each protocol the number of species found at each site, S(obs), was calculated and expressed as a percentage of the total species found at each site for all methods. Jaccard coefficients were used to describe the relationship between the species compositions of sites. This analysis was repeated for the subset of species determined to be threatened because of their restricted ranges. To give an indication of the total number of species across the eight sites, a species area curve, based on the 6 m  50 m protocol, was constructed using pooled data from all sites. The 6 m  50 m protocol was used for this as it was the protocol, which revealed most species. Two non-parametric estimators of species richness, the incidence coverage estimator (ICE) and Chao 2, were used to estimate total species richness across the sites. In a comparative study of several such estimators, Chazdon et al. (1998) showed that these two best satisfied the requirement of a species-richness estimator for a Costa Rican humid tropical forest. Both these estimators work on the assumption that the less common species found within a sample can yield information on the even less common unsampled species. Chao 2 uses the ratio of uniques (species that are found in only one sampling unit) to duplicates (those found in two sampling units) to calculate the number of additional species that are likely to be found outside of the fraction of forest that fell into the sample. ICE is based on the number of uniques and the number of infrequent species, i.e. those species found in less than 10 the total sampling units, regardless of the how those units are defined. Chao 2 has the advantage that a standard deviation can be calculated, and hence a confidence interval for the estimate of total species (Colwell, 2004). Formulas for ICE

567

and Chao 2 are given by Chazdon et al. (1998) and nonparametric estimators are reviewed by Walther and Moore (2005). The ability of the protocols to distinguish between S(obs) at different sites was tested by comparing the mean number of species per plot and confidence intervals after 15 repetitions. Because of the lack of within site repetitions confidence intervals could not be calculated for the ad hoc protocol, and it was therefore excluded from this part of the analysis. All protocols were included in the analysis of conservation priorities in which sites were ranked in importance based on S(obs) and on number of restricted range species. Wilcoxon’s signed rank test (two-tailed) was used for comparison of rankings of pairs of protocols. The total number of species was estimated for each protocol at each site using the two nonparametric estimators, ICE and Chao 2. Jaccard coefficients, confidence intervals and non-parametric estimators were calculated using EstimateS (Colwell, 2004). The efficiency of the protocols used was compared using species accumulation curves, with the x-axis measuring time spent on data collection for each protocol. The time taken to complete each repetition of each protocol was recorded by the data recorder, an activity which added an insignificant amount of time to each operation. In comparing efficiencies, it was assumed that all other activities related to the assessments, such as time spent travelling to each site, were independent of the assessment protocol used. To test the performance of the sampling protocols with respect to complementarity reserve selection, sites were then ranked using the ‘greedy’ heuristic selection algorithm of Briers (2001). This algorithm was chosen because of its simplicity and long history (Margules et al., 1988) and also because many other heuristic algorithms for site prioritisation that are now in use are variants of this basic algorithm (Sarkar, 2004). It selects the most speciose site first, then the site that contains most species not already represented in the first sample, followed by the site with most species not already represented in the first two sites, and so on until all species are represented. The sites were ranked in the order that the algorithm selected them on the principal that sites selected first contribute more new species than sites selected later and are therefore more important to an optimal reserve design. The site rankings were then tested using Spearman’s r. 4. Results 4.1. Summary statistics: species diversity Table 1 shows that site 8, the site determined by visual inspection to be most distinct was revealed by all protocols, except the ad hoc method, to be the least diverse as measured by species observed. Site 6 was revealed to be the most diverse. The ad hoc protocol and the 6 m  50 m protocol made the largest contributions to the total observed species list for each site, with 84.6% and 83.4%, respectively. When sites were ranked in order of S(obs) all combinations of pairs of ranks between the 2  50, the 6  50 and the 15 tree protocols were

568

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573

Table 1 Summary statistics for tree inventories of eight tropical dry forest sites in Oaxaca, Mexico using four inventory protocols Site 1

Site 2

Site 3

Site 4

Site 5

Site 6

Site 7

Site 8

Mean

All sites

Total S(obs)

71

61

80

61

67

96

88

50

71.8

169

2  50 protocol S(obs) % of total S(obs)

48 67.6

46 75.4

47 58.8

38 62.3

48 71.6

49 51.0

44 50.0

21 42.0

42.6 59.8

123 72.8

6  50 protocol S(obs) % of total S(obs)

58 81.7

54 88.5

67 83.8

48 78.7

56 83.6

74 77.1

67 76.1

30 60.0

56.8 78.7

141 83.4

Fixed count S(obs) % of total S(obs)

50 70.1

47 77.0

56 70.0

37 60.7

51 76.1

56 58.3

49 55.7

26 52.0

46.5 65.0

121 71.6

Ad hoc S(obs) % of total S(obs)

43 60.6

34 55.7

61 76.3

50 82.0

50 74.6

76 79.2

66 75.0

44 88.0

53.0 73.9

143 84.6

S(obs), species observed in the given protocol; Total S(obs), number of species observed in the four protocols combined; CI, 95% confidence interval.

found to be significantly different ( p < 0.05, Wilcoxon’s signed rank test) whilst all comparisons with the ad hoc protocol were not significant. Overall, Table 2 also confirms the distinctness of site 8 in species composition. Its mean Jaccard coefficient of similarity with all other sites was 0.15, which was significantly different to the mean of 0.49 of all other between site comparisons (t-test: p < 0.01). The most speciose site, site 6, had a mean Jaccard coefficient of 0.49 (excluding comparisons with site 8) and therefore high diversity in this site was not related to a distinct species composition. This suggests that the importance of this site will vary depending upon whether its importance is measured by species number or by its contribution to complementarity reserve selection. Whilst the ad hoc protocol revealed the greatest number of species, the 6 m  50 m protocol revealed a similar number, and because of the repetitions of plots, it is this latter protocol which is used to derive a pooled total species count and the two pooled estimated species counts derived from ICE and Chao 2 for the area as a whole. These are shown in Fig. 1. Fig. 1 shows that the species observed curve, S(obs), reaches a total of 141, but is not yet at asymptote, and hence it is appropriate that estimators of total species richness are used (Colwell and Coddington, 1994). Here Chao 2 estimates a total species count of 161 and ICE estimates 157, suggesting that for

this pooled sample, the observed number is between 87.6% and 89.8% of the true total. Note that the peaks near the starting points of the two estimators are artefacts of their underlying mathematics. However neither of the two non-parametric estimators have yet to reach asymptote and so further sampling would be required to yield a stable estimate of total species from either. This might be expected given that it is known that in these forests all sampling protocols combined revealed a total of 169 species (Table 1). 4.2. Summary statistics: threatened species Table 3 shows site 8 to be the least diverse in threatened species and site 6 was revealed to be the most diverse. As with total species, the ad hoc protocol and the 6 m  50 m protocol made the largest contributions to the total observed threatened species list for each site, with 85.0% and 87.5%, respectively. When the rank order of sites was compared only the comparisons between the 2  50 and the 6  50 protocols

Table 2 Jaccard coefficients of similarity between eight tropical dry forest sites in Oaxaca, Mexico based on total species found by all protocols for each site

Site Site Site Site Site Site Site

2 3 4 5 6 7 8

Site 1

Site 2

Site 3

Site 4

Site 5

Site 6

0.49 0.47 0.48 0.50 0.49 0.43 0.15

0.51 0.46 0.47 0.5 0.46 0.15

0.57 0.48 0.56 0.54 0.18

0.45 0.43 0.54 0.16

0.54 0.46 0.11

0.53 0.16

Site 7

0.17

Fig. 1. Pooled species area curve and estimates of total species for eight Mexican dry forests using the 6 m  50 m protocol. S(obs), species observed; ICE, incidence coverage estimator of total species number; Chao 2, Chao 2 estimator of total species number.

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573

569

Table 3 Summary statistics for threatened species from eight tropical dry forest sites in Oaxaca, Mexico inventoried by four protocols Site 1

Site 2

Site 3

Site 4

Site 5

Site 6

Site 7

Site 8

Mean

All sites

Total TS(obs)

19

13

22

13

19

2  50 protocol TS(obs) % of total TS(obs)

23

16

6

16.4

40

14 74

10 77

11 50

7 54

12 63

12 52

12 75

5 83

10.4 66

28 70.5

6  50 protocol TS(obs) % of total TS(obs)

16 84

12 92

19 86

8 62

15 79

19 83

16 100

6 100

13.9 86

35 87.5

Fixed count TS(obs) % of total TS(obs)

14 74

9 69

15 68

7 54

13 68

15 65

11 69

5 83

11.1 75

31 77.5

Ad hoc S(obs) % of total TS(obs)

14 74

9 69

14 64

10 77

15 79

17 74

10 63

6 100

11.9 75

34 85.0

Total TS(obs), number of threatened species observed by the four protocols; S(obs), species observed by the given protocol; TS(obs), number of threatened species observed.

and between the 6  50 and 15 tree protocols were found to be significantly different ( p < 0.05, Wilcoxon’s signed rank test). Table 4 also confirms the distinctness of site 8 in threatened species composition. Its mean Jaccard coefficient was 0.10, which is significantly different to the mean of 0.34 of all other between site comparisons (t-test: p < 0.01). These coefficients are reduced in comparison to their equivalents for total species; between site similarity is reduced when comparisons are limited to the 20% of species of most restricted range. 4.3. Differences in S(obs) and non-parametric estimation of total species Fig. 2a–c shows the species observed, S(obs), for each site for each of three protocols and non-parametric estimates of total species given by ICE and Chao 2 (confidence limits cannot be calculated for ICE). Overall, the protocols performed poorly in distinguishing between sites with considerable overlap in the confidence intervals of S(obs) from each site. All three consistently distinguished the low diversity site 8 from all others, but the 2 m  50 m protocol (Fig. 2a) was unable to distinguish between any other sites. The 6 m  50 m protocol (Fig. 2b) performed slightly better; in addition to the distinguishing between site 8 and all others, site 4, the second least species rich site, was separable from sites 1, 3, 6 and 7 and site 2 from site 3. Table 4 Jaccard coefficients of similarity between eight tropical dry forest sites in Oaxaca, Mexico based on threatened species found by all protocols for each site

Site Site Site Site Site Site Site

2 3 4 5 6 7 8

Site 1

Site 2

Site 3

Site 4

Site 5

Site 6

Site 7

0.22 0.42 0.44 0.33 0.36 0.28 0.12

0.35 0.19 0.33 0.37 0.27 0.15

0.60 0.53 0.41 0.26 0.05

0.25 0.29 0.25 0.07

0.33 0.25 0.05

0.42 0.10

0.14

The 15 tree protocol separated site 4 from all other sites, as well as site 8 (Fig. 2c). 4.4. Inventory efficiency Fig. 3 shows the mean inventory efficiency across the eight sites for the four protocols measured by number of species encountered against the mean time the three-person inventory team spent on the protocol. The rate of species accumulation is measured by the gradient of the curve. All four curves start steep when ‘new’ species are rapidly encountered and then flatten off as additional species are encountered less often. This is the typical behaviour of the species accumulation curve in a diverse natural forest. However, none of the curves is near to being asymptotic. By this measure the ad hoc protocol is the most efficient, its gradient is the steepest over the first 70 min of effort, or two thirds of its length. The abrupt flattening out of this curve after approximately 70 min of effort is an artefact of the inventory team being allowed to decide for themselves when the point had been reached at which no more species were likely to be found. The fixed count and the 6 m  50 m protocols trace similar curves and therefore have similar efficiencies. Thus by increasing the number of fixed count plots until the same amount of time had been spent on their enumeration as was spent on the 6 m  50 m plots, a similar S(obs) would be attained. It appears that the extra establishment time required for the 6 m  50 m plot is compensated for by the plots containing more individual trees (a mean of 27.4 per 6 m  50 m plot compared to 15 in the fixed count plot). However, the greater number of repetitions of plots that would result from extra investment in the fixed count protocol would likely result in more precise diversity estimates of species number, that is, it is a more statistically efficient protocol. Despite being the most widely used, the 2 m  50 m protocol is shown to be the least efficient. Its high establishment cost is not compensated by a high number of individual trees per plot. On average only 9.1 trees were found in each plot.

570

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573 Table 5 Ranking of eight dry forest sites in Oaxaca, Mexico by four inventory protocols where rank is determined by the relative contribution of each site to reserve prioritisation by the ‘greedy’ selection algorithm

Site Site Site Site Site Site Site Site

1 2 3 4 5 6 7 8

2 m  50 m

6 m  50 m

15 Tree

Ad hoc

All data pooled

7 8 2 6 5 1 4 3

4 2 8 7 6 1 5 3

5 7 1 8 2 4 6 3

6 7 8 5 4 1 3 2

4 7 5 8 6 1 3 2

4.5. Performance of inventory protocols under complementary reserve selection

Fig. 2. Number of species observed, S(obs), and two non-parametric estimators of species richness for eight dry forest sites in Mexico using the (a) 2 m  50 m, (b) 6 m  50 m and (c) fixed count inventory protocol. ICE, incidence coverage estimator of total species number; Chao 2, Chao 2 estimator of total species number.

Complementarity analysis distinguishes between sites based on species composition not species richness. Whilst noting that Jaccard coefficients typically underestimate similarity between diverse forests when sampling fractions are small (Chao et al., 2005), the differences in species compositions illustrated in Tables 2 and 4 suggest that complementarity analysis might be an appropriate tool for prioritisation amongst these sites. Table 5 shows the results of site prioritisation by the greedy selection algorithm, where priority is determined by the order of the selection. For comparison the algorithm was also run using pooled data from all protocols at each site to give a site selection based on the maximum information available and that therefore can be considered the most likely to be correct. No two protocols resulted in the same order of priority and, at least for this selection algorithm and forest type, complementarity analysis is highly sensitive to the dataset used. There is some consistency notable in Table 5. Site 6 was amongst the first four to be chosen regardless of protocol and site 4 was always amongst the last, however, Table 6 shows that correlations between different protocols, whilst usually positive, were found to be weak, only one being greater than 0.50. By the standard of the ranking of all pooled data, the ad hoc and the 2 m  50 m protocols performed best. Table 7 presents a similar analysis, except that each sample is restricted to the 20% of species considered to be threatened. Again no two protocols resulted in identical rank orders, although sites 1 and 8 were consistently ranked amongst the highest, whilst sites 4 and 2 were consistently ranked amongst the lowest. This is Table 6 Correlation of coefficients (Spearman’s r) of rankings of eight dry forest sites in Oaxaca, Mexico by four inventory protocols where rank is determined by the relative contribution of each site to reserve prioritisation by the ‘greedy’ selection algorithm 2 m  50 m

Fig. 3. Efficiency of four inventory protocols in eight Mexican dry forests.

6 m  50 m

15 Tree

6 m  50 m 15 Tree Ad hoc

0.00 0.62 0.50

0.21 0.50

0.02

All data pooled

0.69

0.52

0.38

Ad hoc

0.69

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573 Table 7 Ranking of eight dry forest sites in Oaxaca, Mexico by four inventory protocols where rank is determined by the relative contribution of threatened species made by each site to reserve prioritisation by the ‘greedy’ selection algorithm

Site Site Site Site Site Site Site Site

1 2 3 4 5 6 7 8

2 m  50 m

6 m  50 m

15 Tree

Ad hoc

All data pooled

1 – 5 – 6 3 2 4

3 – 1 – 6 2 5 4

2 6 1 7 – 4 5 3

4 7 – 5 2 1 6 3

2 5 3 – – 1 6 4

Sites not selected are those whose complement of threatened species are completely represented by other sites.

reflected in the generally stronger correlations shown in Table 8. Note that the ad hoc protocol performed badly compared to the pooled ranking. 5. Discussion Given that none of the protocols successfully distinguished between sites based on S(obs), it is suggested that more intensive sampling would be required if these protocols were to be used to select reserves based on species number. Given that even the least intensive methodology employed here, the 2 m  50 m protocol, was applied more intensively than it is typically (15 instead of 10 repetitions) the suitability of this protocol to make distinctions between sites of similar forest type must be questioned. The ad hoc protocol was shown to be the most efficient measured by the rate of accumulation of new species during sampling, the 6 m  50 m protocol and the 15 tree protocols were similar but less efficient and the 2 m  50 m the least efficient protocol, suggesting that more intensive application of the 2 m  50 m protocol would be the least efficient way of increasing the sampling intensity. The Chao 2 estimator is shown to give a significantly different estimate of total species per site to S(obs) for seven of the sites sampled by the 2 m  50 m protocol, seven of the sites sample by the fixed count plots but only three of the sites when sampled by the 6 m  50 m protocol. The confidence intervals associated with the Chao 2 estimates are smaller that those of the S(obs), suggesting that prioritisation between sites based on Chao 2 estimates of species richness may be easier than when Table 8 Correlation coefficients (Spearman’s r) of rankings of complement of threatened species from eight dry forest sites in Oaxaca, Mexico by four inventory protocols where rank is determined by the relative contribution of each site to reserve prioritisation by the ‘greedy’ selection algorithm 2 m  50 m

6 m  50 m

6 m  50 m 15 tree Ad hoc

0.64 0.59 0.27

0.85 0.10

0.24

All data pooled

0.60

0.82

0.80

15 Tree

Ad hoc

0.16

Sites not selected by the algorithm are ranked last (8th or 7.5th) in this analysis.

571

based on S(obs). It is to be expected that S(obs) would be lower for the protocols that captured a smaller sampling fraction. The smallest sampling fraction, measured by trees sampled, was captured by the 2 m  50 m protocol with a mean of 136.5 trees per site, followed by the 15 tree protocol with exactly 225 trees per site with the largest being captured by the 6  50 m protocol with a mean of 411 trees per site. If non-parametric estimators are to be useful in these circumstances they should make a proportionately greater addition to the S(obs) derived from smaller sampling fractions so that they more nearly approach the true species richness. Although for some sites (5, 6 and 8) ICE and Chao 2 do greatly increase the estimates derived from the 2 m  50 m protocol, in other cases (sites 3, 4 and 7) they remain below the S(obs) derived from the 6 m  50 m protocols. In other words, these non-parametric estimates must be incorrect and thus their usefulness in site prioritisation must be questioned (Chiarucci et al., 2003), at least when based on small samples. Jaccard coefficients can over emphasise differences between vegetation types when sample sizes are small, however, by pooling the results from all protocols for each site this bias is reduced and therefore the between site differences they suggest reflect real differences in species compositions. Most pairs of sites had coefficients of less than 0.5, whether all species or threatened species were considered. This suggests an advantage of complementary reserve selection over prioritisation by species richness, the former uses more information and in this case that information better distinguishes the sites. Just as there were significant differences between site priorities based on richness, the samples derived from the four protocols also suggested different priorities when subjected to the greedy selection algorithm. Consideration therefore needs to be given to the sensitivity of reserve selection procedures to data quality before it can be claimed that a prioritisation exercise provides an optimal solution to selecting reserve networks. As it must be assumed that the protocols that capture the highest number of species, or threatened species, most nearly approximate the true species number for each site, complementary analysis based on higher S(obs) and higher numbers of threatened species can be assumed to more closely approach most efficient reserve network solutions. Thus the ad hoc and 6 m  50 m protocols would be preferable in complementarity analysis. As well as their giving a different order of priorities, prioritisation based on threatened species was also notably different in that fewer sites were required to conserve all threatened species than were required to conserve all species, evidence that conservation directed only towards threatened species is likely to require fewer resources. This is simply a reassertion of the hotspot principal of Myers et al. (2000) but applied at a local, rather than a global scale. Compared to selecting sites by their relative number of species, which would result in the least speciose site, site 8 always being determined as the least important site, the greedy algorithm selected it as one of most important sites under all protocols, its importance being that it contains a high

572

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573

proportion of species not found at other sites. It is therefore probable that in selecting between more clearly differentiated sites, this algorithm would be less sensitive to the dataset used and hence produce more consistent rank orders. The recommendations that follow from this work are therefore as follows. The ad hoc protocol was highly efficient and simple. Where statistical analysis is not considered essential, it may be a suitable protocol if resources are especially limited. The control of effort between samples is the fundamental problem with ad hoc or checklisting methods of biodiversity assessment. Controlling effort by time may offer a means of ensuring comparability between samples, however, this is not simple, as was shown by the abrupt flattening out of the ad hoc curve in Fig. 3. The 6 m  50 m and the 15 tree protocols are more statistically robust, and relatively efficient, and therefore merit consideration as inventory protocols. Where the objectives of local prioritisation of sites and comparison with inventories published elsewhere need to be met, then the 6 m  50 m protocol may have much to offer. A notional 2 m  50 m plot can be extracted from the data from each 6 m  50 m plot providing the data recorder notes which individual trees fall within 1 m either side of the central 50 m line around which the 6 m wide plot is established. Thus, better within locality sampling can be achieved without compromising between locality comparisons. The recommendations that arise from this work are necessarily generalized as the appropriateness of each protocol for a given assessment will depend on many variables, which could not be tested here. These include the precise objectives of the assessment, the forest type under investigation and the delineation of the sampled population (here all woody stems above 5 cm dbh). Higgins and Ruokolainen (2004) showed that very different efficiencies resulted when the sampled population was delineated in different ways whilst the sampling protocol was kept constant. However, we emphasise that based on the analysis presented, the 2 m  50 m protocol, variants of which are amongst the most widely used in published floral surveys, has little to recommend it other than that surveys resulting from it can be directly compared with similar surveys. The lesson from this work is that the appropriateness of variants of this protocol, or indeed any other, cannot be assumed, a priori, to be either efficient or effective in distinguishing between the forest types to be assessed. We suggest that for locally based organisations efficiency and effectiveness may be of far greater importance than the ability to compare local results with other published studies from distant localities. Acknowledgements This research was funded by the UK Government’s Economic and Social Research Council and Natural Environment Research Council (R42200134147) and by the European Commission as part of the BIOCORES Project (PL ICA4-200010029). We thank the Mexican NGO GAIA A.C. for their support during completion of fieldwork, and two anonymous reviewers for helpful comments on an earlier manuscripts.

References Balmford, A., Jayasuriya, A.H.M., Green, M.J.B., 1996. Using higher-taxon richness as a surrogate for species richness. II. Local applications. Proc. Roy. Soc. Lond. B: Biol. Sci. 263, 1571–1575. Briers, R.A., 2001. Incorporating connectivity into reserve selection procedures. Biol. Conserv. 103 (1), 77–83. Cantu´, C., Wright, R.G., Scott, J.M., Strand, E., 2004. Assessment of current and proposed nature reserves of Mexico based on their capacity to protect geophysical features and biodiversity. Biol. Conserv. 115, 411–417. Ceballos, G., Rodrı´guez, P., Medellı´n, R.A., 1998. Assessing conservation priorities in megadiverse Mexico: mammalian diversity, endemicity and endangerment. Ecol. Appl. 8, 8–17. Chao, A., Chazdon, R.L., Colwell, R.K., Shen, T.-J., 2005. A new statistical approach for assessing similarity of species composition with incidence and abundance data. Ecol. Lett. 8, 148–159. Chase, M.K., Kristan, W.B.I., Lynam, A.J., Price, M.V., Rotenberry, J.T., 2000. Single species as indicators of species richness and composition in California costal sage scrub birds and small mammals. Conserv. Biol. 14, 474– 487. Chazdon, R.L., Colwell, R.K., Denslow, J.S., Guariguata, M.R., 1998. Statistical methods for estimating species richness of woody regeneration in primary and secondary rain forests of northeastern Costa Rica. In: Dallmeier, F., Comiskey, J.A. (Eds.), Forest Biodiversity Research, Monitoring and Modeling: Conceptual Background and Old World Case Studies. UNESCO/Parthenon Publishing Group, New York, USA, pp. 285– 309. Chiarucci, A., Enright, N.J., Perry, G.L.W., Miller, B.P., 2003. Performance of non-parametric species richness estimators in a high diversity plant community. Divers. Distrib. 9, 283–295. Colwell, R.K., 2004. EstimateS 7 User’s Guide, Online. University of Connecticut, Storrs. , http://www.viceroy.eeb.uconn.edu/EstimateS7Pages/EstS7U sersGuide/EstimateS7UsersGuide.htm. Colwell, R.K., Coddington, J.A., 1994. Estimating terrestrial biodiversity through extrapolation. Phil. Trans. Roy. Soc. Lond. B 345, 101–118. Condit, R., Foster, R., Hubbell, S.P., Sukumar, R., Leigh, E.G., Manokaran, N., de Lao, S.L., LaFrankie, J.V., Ashton, P.S., 1998. Assessing forest diversity on small plots: calibration using species-individual curves from 50-ha plots. In: Dallmeier, F., Comiskey, J.A. (Eds.), Forest Biodiversity Research, Monitoring and Modeling: Conceptual Background and Old World Case Studies. UNESCO/Parthenon Publishing Group, New York, USA, pp. 247– 268. Davis, S.D., Heywood, V.H., Herrera-MacBryde, O., Villa-Lobos, J., Hamilton, A.C. (Eds.), 1997. Centres of Plant Diversity. A Guide and Strategy for Their Conservation. WWF/IUCN, Gland, Switzerland. Dinerstein, E., Olson, D.M., Graham, D.J., Webster, A.L., Primm, S.A., Bookbinder, M.P., Ledec, G., 1995. A Conservation Assessment of the Terrestrial Ecoregions of Latin America and the Caribbean. World Bank/WWF, Washington, DC. Droege, S., Cyr, A., Larive´e, J., 1998. Checklists: an under-used tool for the inventory and monitoring of plants and animals. Conserv. Biol. 12, 1134– 1138. Dudley, N., Jeanrenaud, J.-P., 1998. Needs and prospects for international cooperation in assessing forest biodiversity: an overview from WWF. In: Bachmann, P., Ko¨hl, M., Pa¨ivinen, R. (Eds.), Assessment of Biodiveresity for Improved Forest Planning. Kluwer Academic Publisher, Dordrecht, Netherlands, pp. 31–41. Gaston, K.J., Rodrigues, A.S.L., 2003. Reserve selection in regions with poor biological data. Conserv. Biol. 17, 188–195. Gentry, A.H., 1982. Patterns of neotropical plant species diversity. Evol. Biol. 15, 1–84. Gordon, J.E., 2006. The role of science in NGO mediated conservation: insights from a biodiversity hotspot in Mexico. Environ. Sci. Policy 9, 547– 554. Gordon, J.E., Hawthorne, W.D., Reyes-Garcı´a, A., Sandoval, G., Barrance, A.J., 2004. Assessing landscapes: a case study of tree and shrub diversity in the seasonally dry tropical forests of Oaxaca, Mexico and southern Honduras. Biol. Conserv. 117, 429–442.

J.E. Gordon, A.C. Newton / Forest Ecology and Management 237 (2006) 564–573 Gordon, J.E., Gonza´lez, M.-A., Herna´ndez, J.V., Lavariega, R.O., ReyesGarcı´a, A., 2005. Guaiacum coulteri: an over-logged dry forest tree of Oaxaca, Mexico. Oryx 39, 82–85. Gordon, J.E., Bowen-Jones, E., Gonza´lez, M., 2006. What determines dry forest conservation in Mesoamerica? Opportunism and pragmatism in Mexican and Nicaraguan protected areas. In: Pennington, R.T., Ratter, J.A., Lewis, G.P. (Eds.), Neotropical Savannas and Dry Forests: Plant Diversity, Biogeography and Conservation. CRC-Taylor Francis, Boca Raton, USA, pp. 343–357. Gordon, J.E., Newton, A.C., 2006. The potential misapplication of rapid plant diversity assessment in tropical conservation. J. Nat. Conserv. 14 (2), 117–126. Hall, J.B., 1991. Multiple-nearest-tree sampling in an ecological survey of Afromontane catchment forest. Forest Ecol. Manage. 42, 245–299. Hawthorne, W.D., Abu-Juam, M., 1995. Forest Protection in Ghana. IUCN, Gland, Switzerland and Cambridge, UK. Heywood, V.H. (Ed.), 1995. Global Biodiversity Assessment. UNDP/CUP, Cambridge. Higgins, M., Ruokolainen, K., 2004. Rapid tropical forest inventory: a comparison of techniques based on inventory data from Western Amazonia. Conserv. Biol. 18, 799–811. Hilton-Taylor, C., Mace, G.M., Capper, D.R., Collar, N.J., Stuart, S.N., Bibby, C.J., Pollock, C., Thomsen, J.B., 2000. Assessment mismatches must be sorted out: they leave species at risk. Nature 404, 541. Kerr, J.T., Sugar, A., Packer, L., 2000. Indicator taxa, rapid biodiversity assessment, and nestedness in an endangered ecosystem. Conserv. Biol. 14, 1726–1734. Lawrence, A., 2002. Participatory assessment, monitoring and evaluation of biodiversity: summary of the ETFRN internet discussion 7–25 January 2002. In: Vol. http://www.etfrn.org/etfrn/workshop/biodiversity/index.html (February 2004). Margules, C.R., Pressey, R.L., 2000. Systematic conservation planning. Nature 405, 243–252. Margules, C.R., Nicholls, A.O., Pressey, R.L., 1988. Selecting networks of reserves to maximize biological diversity. Biol. Conserv. 43, 63–76. Myers, N., Mittermeier, R.A., Mittermeier, C.G., da Fonseca, G.A.B., Kent, J., 2000. Biodiversity hotspots for conservation priorities. Nature 403, 853–858. Phillips, O., Miller, J.S., Gentry, A.H., 2002. Global patterns of plant diversity: Alwyn H. Gentry’s forest transect data. Monographs in Systematic Botany from the Missouri Botanical Garden 89, 1–319. Phillips, O., Martı´nez, R.V., Vargas, P.N., Monteagudo, A.L., Zans, M.-E.C., Sa´nchez, W.G., Cruz, A.P., Timana´, M., Yli-Halla, M., Rose, S., 2003.

573

Efficient plot-based floristic assessment of tropical forests. J. Trop. Ecol. 19, 629–645. Prendergast, J.R., Quinn, R.M., Lawton, J.H., 1999. The gaps between theory and practice in selecting nature reserves. Conserv. Biol. 13, 484–492. Pressey, R.L., 1994. Ad hoc reservations: forward or backward steps in developing representative reserve systems? Conserv. Biol. 8, 662–668. Pressey, R.L., Humphries, C.J., Margules, C.R., Vane-Wright, R.I., Williams, P.H., 1993. Beyond opportunism: key principles for systematic reserve selection. Trends Ecol. Evol. 8, 124–128. Rodrigues, A.S.L., Gaston, K.J., 2002. Maximizing phylogenetic diversity in the selection of networks of conservation areas. Biol. Conserv. 105, 103–111. Royal Society, 2003. Measuring biodiversity for conservation. Royal Society Policy Document (November 2003). Sagar, R., Raghubanshi, A.S., Singh, J.S., 2003. Tree species composition, dispersion and diversity along a disturbance gradient in a dry tropical forest region of India. Forest Ecol. Manage. 186, 61–71. Salafsky, N., Margoluis, R., Redford, K.H., Robinson, J.G., 2002. Improving the practice of conservation: a conceptual framework and research agenda for conservation science. Conserv. Biol. 16, 1469–1479. Salas-Morales, S.H., Saynes-Va´quez, A., Schibli, L., 2003. Flora de la costa de Oaxaca: lista florı´stica de la regio´n de Zimata´n. Boletı´n de la Sociedad Bota´nica Mexicana 72, 21–58. Sarkar, S., 2004. Conservation biology. In: Zalter, N. (Ed.), Stanford Encyclopedia of Philosophy. Summer 2004 ed. The Metaphysics Research Lab, Stanford University, Stanford. , In: http://www.plato.stanford.edu//archives/ sum2004/entries/conservation-biology/. Schulenberg, T.S., Short, C.A., Stephenson, P.J. (Eds.), 1999. A biological assessment of Parc National de la Marahoue´. Coˆte d’Ivoire Conservation International, Washington. Stern, M.J., 1998. Field comparisons of two rapid vegetation assessment techniques with permanent plot inventory data in Amazonian Peru. In: Dallmeier, F., Comiskey, J.A. (Eds.), Forest Biodiversity Research, Monitoring and Modeling: Conceptual Background and Old World Case Studies. UNESCO/Parthenon Publishing Group, New York, pp. 269–283. UNEP-WCMC, 2003. Biodiversity Assessment and Monitoring. Guidance for Practitioners. UNEP0-WCMC, Cambridge, UK. Vanclay, J.K., Gillison, A.N., Keenan, R.J., 1997. Using plant functional attributes to quantify site productivity and growth patterns in mixed forests. Forest Ecol. Manage. 94, 149–163. Walther, B.A., Moore, J.L., 2005. The concepts of bias, precision and accuracy, and their use in testing the performance of species richness estimators, with a literature review of estimator performance. Ecography 28, 815–829.