Biodiversity and Conservation 13: 2599–2621, 2004.

# 2004 Kluwer Academic Publishers. Printed in the Netherlands.

Mapping the spatial distribution of plant diversityindices in a tropical forest using multi-spectralsatellite image classification and field measurements

J. LUIS HERNANDEZ-STEFANONI1 and RAUL PONCE-HERNANDEZ1,2,*1WatershedEcosystemsGraduateProgram,TrentUniversity,Peterborough,Ontario,Canada;2Environmental

andResourceStudiesProgram,DepartmentofGeography,TrentUniversity,Peterborough,Ontario,Canada

K9J 7B8; *Author for correspondence (e-mail: rponce@trentu.ca; fax: þ1-705-748-1011-extn. 1569)

Received 12 November 2002; accepted in revised form 8 September 2003

Key words: Biodiversity, a- and b-diversity, Landscape, Mapping, Remote sensing, Satellite image

classification, Tropical forest

Abstract. The relationships among alpha and beta diversity indices, computed from 141 randomly

sampled quadrats, and the vegetation classes obtained by multi-spectral satellite image classification,

were used as a strategy for mapping plant diversity in a tropical landscape mosaic. A relatively high

accuracy of the land cover map was revealed by the overall accuracy assessment and the Cohen’s Kappa

statistic. Species accumulation models were used to evaluate how representative the sample size was of

the different vegetation types. A standard one-way, between-subjects ANOVA confirmed a significant

reduction of the within-class variance of plant diversity with respect to their total variance across the

landscape. Computed uniformity indices, to assess the internal uniformity of vegetation classes on the

diversity indices, confirmed the goodness of the mapped classes in stratifying variability of plant di-

versity. This allowed for the use of the mapped classes as spatial interpolators of plant diversity values

for estimation and up-scaling purposes. Finally, it was revealed that the plant diversity of the landscape

depends, to a large extent, on the diversity contained in the most mature forest class, which is also the

most diverse community in the studied area. High and moderate beta diversity values between mature

forests and both the secondary associations and the first stages of succession, respectively, indicated that

there is a significant contribution to the diversity of the landscape by those vegetation classes.

Introduction

Tropical forests of the Yucatan peninsula in Mexico consist not only of mature forests

but also of other habitats (Cabrera et al. 1982). In the context of the landscape, they

form a mosaic of forests in different stages of succession, which together with other

non-forested land cover habitats, such as flooded or deforested areas, croplands and

grasslands, produce a relatively broad range of ecological conditions. These habitats

vary in size, shape, and composition. These variations are attributed to a combination

of interacting physical, biological and anthropogenic elements (Forman 1995). In

some cases the disturbance regime is dominated by natural causes, for example fire

and hurricanes (Whigham et al. 1991). In others, disturbances are of anthropogenic

origin, such as slash-and-burn shifting cultivation and land use changes. The char-

acteristics, spatial distribution and configuration of these habitats may also influence

the presence and abundance of species (Mazerolle and Villard 1999), causing them to

be unevenly distributed over space.

During the past decade the forests of the Yucatan peninsula have endured high

rates of deforestation (SEMARNAP 1998). The resulting effects of this process are

the loss of biological diversity and damage to wilderness habitats, increase in soil

erosion, disturbance to the hydrological cycle and nutrient losses, among others

(Isik et al. 1997). To stop these processes and preserve the biological diversity of

these forests, accurate information on their biodiversity is required. Developing and

using this type of information is therefore an essential part of conservation pro-

grams. For example, the spatial distribution of species over a landscape aids in the

identification of high priority areas (Carroll 1998; Myers et al. 2000), which may be

used for locating reserves, refuges or other protected areas.

One approach to describing the spatial patterns of plant diversity in landscape

mosaics consists of accounting for the diversity within and between particular

habitats that could be considered homogeneous communities. On the one hand,

plant diversity is estimated inside each of these habitats. On the other, the dis-

similarity (or complementarity) between such habitats is also estimated (Colwell

and Coddington 1995). Whittaker (1972) defined these two components of species

diversity as a- and b-diversity (i.e., within and between communities). This ap-

proach to measuring plant diversity is useful in assessing not only the relative

importance in terms of diversity of different areas, but also in estimating dissim-

ilarities in species composition among habitats or ecosystems in a landscape. The

‘intrinsic’ diversity of a community is given by its a-diversity. Thus, an area with

higher a-diversity may be considered more important than one with lower a-di-

versity values, for conservation purposes. The contribution of any given area or

habitat to the overall diversity in the landscape can be determined through the

estimation of b-diversity. In consequence, areas with lower a-diversity can still be

of importance for conservation because of their contribution to the total diversity at

the landscape scale (Groombrige 1992). That is to say, there are some species that

are only present in these particular areas.

The most commonly employed measures for estimating the diversity of species

in a community are those related to species richness (i.e., the number of species

present in an area) and measures based on species frequencies or abundance, in-

cluding Shannon and Simpson indices (Magurran 1988; Krebs 1989). Several

measures of species diversity among communities have been recommended to

assess biodiversity through environmental gradients (Whittaker 1972; Magurran

1988). These measures, however, can also be applied to assess the replacement of

species among habitats in a heterogeneous landscape mosaic (Moreno and Halffter

2001).

One of the main problems in comparing the number of species among com-

munities is that species richness is not independent of the sample size. The number

of species increases with the size of the area sampled. Therefore, to make com-

parable the number of species among different habitat types, it is necessary to

employ the same sampling effort in every one of them (Soberon and Llorente 1993;

Moreno and Halffter 2000). Another problem in measuring biological diversity is

presented by the difficulties and effort required for sampling large areas of densely

forested landscapes, particularly when access to some particular sites is difficult,

2600

which is the case of most tropical forests (Fuller et al. 1997). However, combining

ground surveys with the support from remote sensing image analysis has proven to be

a very useful tool for solving the numerous practical problems involved in this type of

undertaking (Innes and Koch 1998; Nagendra and Gadgil 1999; Muldavin et al. 2001;

Kerr and Ostrovsky 2003). So, multi-spectral satellite images can be used for iden-

tifying and mapping land cover classes. Such classes can be taken, in a broad sense,

as being equivalent to habitats. Mapping such classes offers several advantages in the

assessment of biodiversity over the landscape. First, the diversity within the mapped

classes can be assessed relatively easy through field measurements. Second, land

cover classes could be sufficiently linked to species composition and abundance in

those particular areas over the landscape (Nagendra and Gadgil 1999).

This paper introduces an approach to the assessment of plant diversity of tropical

forests and for mapping the spatial distribution of plant diversity indices across the

landscape. The approach is based on procedures that use satellite image classifi-

cation, supported by ground observations and measurements, to map out land cover

classes and to use this map for both description and up-scaling of ground mea-

surements and indices of plant diversity across the landscape mosaic. In the context

of this study, plant diversity is understood as the composition of trees, shrubs and

vines sampled during field survey in the area. Such measurements helped in

computing a- and b-diversity indices. The technical issues examined in this paper

are related to the process of land cover map production, assessing the completeness

of the sample size for the field survey, and exploring the relationships between the

mapped classes and diversity indices, so that the classes can be used as part of an

integrated framework for up-scaling estimates and for the analysis of the spatial

patterns of plant diversity of the area of concern.

Methods

Study area

This study was conducted in the State of Quintana Roo, on the eastern seaboard of

the Yucatan peninsula, southeastern Mexico. The study area is located near the

village of Buena Vista, on the western shore of Bacalar lagoon (18853054@–18858014@ N latitude and 88810004@–88814037@ W longitude) and covers 64 km2

(8 km� 8 km) (Figure 1). The climate of the area is tropical warm sub-humid with a

dry period (i.e., Climate type Aw). The mean annual temperature is 26 8C, the coldest

month is January with 23 8C and the warmest period is between July and August with

29 8C. The annual rainfall is between 1000 and 1300 mm, concentrated in a period

between June and October, whereas the dry season is from December to April

(Cabrera et al. 1982). The predominant soil orders (as per the FAO-UNESCO soil

classification) present in the study area are Rendzinas Lithosols, Chromic Luvisols

and Cambisols, with small inclusions of Gleyic Vertisols and Gleysols (Duch 1995).

The landscape is dominated by tropical sub-deciduous forests with two or three

levels of canopy, consisting of strata of trees and woody creeping plants between 7

2601

and 25 m in height. Strata of shrubs and herbs made up by re-growth, saplings,

succulents and epiphytes are also part of the vegetation. In this type of forest, about

75% of all tree species are evergreen with leaves all year round. The most common

species include Brosimun alicastrum, Bursera simaruba, Manilkara zapota,

Metopium brownei, Psidium sartorium and Vitex gaumeri. There are also secondary

plant associations in the study area. Among them are ‘savanna’, distinguished for

having few sparse tree species between 3 and 10 m in height and a grassy and herb

stratum dominated by Scleria pterota. The other important secondary association is

‘Akalche’ (in local Mayan language) consisting of a shrub stratum dominated by

Hemotoxylon campechianum. Both of these plant associations are found in flooded

areas or zones with deficient drainage (Cabrera et al. 1982).

The tropical sub-deciduous forests of the study area have endured disturbances of

different intensity, such as incoming hurricanes from the Caribbean Sea, ‘slash and

burn’ agricultural practices and other natural and anthropogenic disturbances,

which have changed the structure of the vegetation. Four different stages of suc-

cession, characterized by their age, can be found in the area. Local people have

developed names for these stages, using mainly the Mayan language. So, ‘kanah

kax’ refers to a forest of 20 or more years of age; ‘kelenche’ is used for vegetation

between 11 and 19 years of age; ‘juche’ is used for plant species between 4 and 10

years of age and ‘saakab’ for plant species of 3 or less years of age. These terms

have a deeply rooted ecological meaning in the local indigenous knowledge, related

to the use and management of tropical forests for practical purposes.

Figure 1. Location of the study area.

2602

Land cover mapping

A false colour composite image was used for mapping land cover classes in

the study area. The image was created using bands 5 (short-wave infrared:

1.55–1.75 nm), 4 (near infrared: 0.76–0.90 nm) and 3 (red: 0.63–0.69 nm) of the

Landsat 7 Thematic Mapper (TM) imagery acquired in April 2000. Every band

was already geo-referenced and radiometrically corrected. This image also served

as a spatial reference framework for locating suitable ‘training’ sites of all land

cover types to be identified on the ground.

At least two ‘training sites’ were selected and located within each land cover

type. These areas had a minimum of 1.0 ha of extent. Their exact location on the

ground was determined using a hand-held Global Positioning System (GPS) unit.

The training sites were identified and located in the field within the following land

cover types: Kanah kax, kelenche, juche, saakab, akalche, savanna, deforested

areas, grasslands and cropping areas. Water bodies were identified in the false

colour composite image. After removing some useless or redundant training sites, a

total of 26 ‘training’ areas were used for producing a land cover map.

In addition to the ‘training’ sites, six spectral bands (TM 1–5 and 7) of Landsat

TM 7 imagery were used in the supervised classification routine, to produce a land

cover raster map. The Maximum Likelihood Algorithm as implemented by ER

MapperTM 6.1 (Earth Resource Mapping Ltd. 1998) was used to perform the

supervised classification. After several trials removing some of the bands, a final

land cover raster map was produced using red, near infrared and short-wave in-

frared bands (TM 3–5). To remove the speckle in the classified image, two filters

were applied. The first one was a 3� 3 median filter to smooth out the image for

continuity appearance. The second filter applied, known as ‘majority’ filter, was

also used for smoothing the classified image (Earth Resource Mapping Ltd. 1998).

One hundred and forty one (141) randomly selected field-sampling sites were

used for assessing the accuracy of the classified land cover map. These samples had

a dual function; they were used not only to compute plant diversity indices at

sampling locations, but also to test the accuracy of the classes generated from the

classification of the satellite FCC image. Two procedures were used for accuracy

assessment, namely, overall accuracy and Cohen’s Kappa statistic. These are the

most commonly used techniques for assessing the agreement between classes in the

map and ‘true’ classes on the ground (Campbell 1987). The overall accuracy of the

classified image was computed by dividing the total number of correctly classified

sample units verified on the ground, over the total number of reference sample

units. The overall accuracy (ACo) can be written as:

ACo ¼ SQcc

SQtot

where SQcc is the total number of sample units correctly classified and SQtot

represents the total number of reference sample units placed in the field.Cohen’s Kappa statistic (Campbell 1987) measures the difference between

the observed agreements among the classes on the ground and those in the map, and

2603

the agreement that might be obtained by chance alone. The equation of the Kappa

statistic can be written as:

K ¼ O � E

1 � E

where ‘O’ is the overall proportion of corrected classified types in the image,

calculated as the sum of diagonal entries divided by the total number of samples,

and ‘E’ is an estimate of the chance agreement to the observed percentage correct.

These values are calculated using the row and column totals as the sum of the

products of row and column totals in the diagonal divided by the grand total of

these products. The kappa (K) coefficient will equal 1 if there is perfect agreement,

whereas 0 is what would be expected by total chance alone. The K statistic can be

interpreted as the accuracy with which any classified map unit would predict the

corresponding land cover class on the ground.

Field sampling design

The biological composition (plant diversity) of several vegetation types was esti-

mated by data obtained from a field survey. Plant surveys were performed in two

stages at different time periods, the first during June and July of 2000 and the other

during July and August of 2001. The sample unit on the ground consisted of a site

of two nested quadrats. One of them, of size 10 m� 10 m, was used to sample tree

species and vines from various identifiable forest canopy levels. The first canopy

layer consisted of trees that have a height of 20 m or more. In the second canopy

layer, trees between 10 and 20 m of height were considered. The third layer con-

sisted of trees with less than 10 m of height. All these trees were identified and

counted. Every one of these quadrats contains a nested sub-quadrat of 5 m� 5 m for

sampling all the shrubs taller than 1.0 m. Consequently, the biological composition

of plants in this study is characterized by trees, shrubs and vines that fell within the

sampling quadrats.

A total of 141 sample sites (quadrats) were placed in six vegetation types using a

stratified random sampling design. Of the total number of quadrats, 42 fell within

the class ‘kanah kax’, 25 in ‘kelenche’, 20 in ‘juche’, 20 in ‘saakab’, 17 in ‘akalche’

and 17 in the ‘savanna’ vegetation class (Figure 2). The samples for every one of

the vegetation types were randomly located on the ground utilizing a GPS unit. All

sample sites, after being validated, were entered into an ad hoc Data Base System

designed, developed and implemented to process the field data.

One of the most difficult tasks during fieldwork was the identification of species

on the ground. A major constraint was the inability to collect plants with all

morphological components needed for species identification in a herbarium. Time

of collection (seasonality) and logistical constraints hampered the effort. Instead,

the knowledge of local people, who have worked in the forest for generations and

who have the ability to identify species using local names, was used as the only

alternative available in the field for plant species identification. Published work

2604

describes the vegetation of the Yucatan peninsula with different degrees of detail.

These sources were consulted to translate Mayan names to scientific species names.

Among these publications are those of Cabrera et al. (1982), Sosa et al. (1985) and

Sarukhan and Pennington (1998).

A total of 182 species were recorded while sampling in the field. Of them, 89.6%

were identified with both local and scientific names, 1.6% were only identified

with local names and the rest, 8.8%, were recognized as different to other species in

the field but they have neither been identified with a local name nor with a scientific

name.

Sample size

A main concern on biodiversity assessments is the number of samples required to

estimate reliably the species diversity of each community in a heterogeneous area

that allows for comparisons of biodiversity among those communities. Two

mathematical models of species accumulation were employed to evaluate the

completeness of inventories of trees, shrubs and vines species in this study. These

models were applied in every one of the six vegetation types and in the landscape as

a whole. The approach uses species accumulation curves to fit models for extra-

polating the number of species. Alternatively, the number of quadrats required to

Figure 2. Location of the sample sites in the studied area for the six vegetation types.

2605

reach a given number of species or a proportion of the number of species estimated

by these models could be calculated (Soberon and Llorente 1993).

Species accumulation curves were computed using number of plant species as a

cumulative variable against number of quadrats as sampling unit. In such curves,

the order in which the samples are added to the total is relevant to the shape of the

curve. To eliminate this effect, the order of entry of the samples was randomized

100 times using EstimateS 5.0.1 software (Colwell 1997). This procedure creates

smoothed species accumulation curves. Such curves were used to fit two asymptotic

species accumulation models using nonlinear regression routines. These models

were reviewed by Soberon and Llorente (1993) and have been applied to assess

completeness of inventories of bats (Moreno and Halffter 2002) and vascular plants

(Miller and Wiegert 1989).

The first, linear dependence (LD) model is recommended for small areas or in

situations where the asymptote can be reached with a finite number of quadrats

(Soberon and Llorente 1993). It has the following expression:

SðqÞ ¼ a

b½1 � expð�bqÞ

where S(q) is the number of species estimated with a number of ‘q’ quadrats, the

parameter ‘a’ represents the rate of increase at the beginning of sampling, while

a=b is the asymptote or number of species estimated. Lamas et al. (1991) re-

commended an equation for estimating the number of samples needed to calculate a

proportion of the total plant species as predicted by the asymptote:

qp ¼ �1

b Lnð1 � pÞ

where ‘p’ is proportion of the total number of species for which the required

number of quadrats is estimated.

The second model, or ‘Clench’ model, is recommended for large areas or where

the probability of adding new species will increase as a larger number of quadrats is

sampled, until a limit number of species is reached (Soberon and Llorente 1993).

This model is defined as:

SðqÞ ¼ aq

1 þ bq

The equation by Soberon and Llorente (1993) was used to calculate a proportion of

the plant species for which the required number of quadrats is estimated:

qp ¼ p

bð1 � pÞ

Measures of plant diversity

The diversity inside a community (a-diversity) was analyzed in two ways: the total

number of species recorded in each habitat type (vegetation class) and the mean

2606

diversity value of the quadrats inside each vegetation type. To measure a-diversity,

species richness, Shannon and Simpson indices were used (Magurran 1988; Krebs

1989).

To compare diversity levels or dissimilarities among the six vegetation types and

to measure the changes in diversity composition of different sampling quadrats

within a vegetation type, Whittaker’s b-diversity index (Whittaker 1972) and Jac-

card’s similarity measure (Magurran 1988) were chosen. Whittaker’s index is

calculated dividing the total number of species in the area of concern by the mean

a-diversity of that area:

�w ¼ S

�� 1

where ‘S’ is the total number of species recorded in the area, and ‘a’ denotes the

average sample a-diversity. The Jaccard measure is one of the most useful measures

of similarity and it uses the number of species of the sites that will be compared for

its calculations:

Cj ¼j

a þ b � j

where ‘j’ is the number of species found in both sites, ‘a’ is the number of species

in the first site and ‘b’ is the number of species in the second site. To compare the

values of this index with those of Whittaker, the inverse (1�Jaccard measure) was

used. Thus, these two indices, Whittaker and Inverse Jaccard, have a minimum

value of 0 when the two compared habitats are identical, and a maximum value of 1

when the two habitats are entirely different.

Relating the mapped vegetation classes to biodiversity indices

The measurements of plant diversity are confined to a discrete number of sampling

sites within an area. However, for managers, this information ought to be on a

spatially continuous basis across the landscape. So, the practical objective of

linking field quadrat measurements of plant diversity to vegetation classes is to

explore the relationships between them with an aim of being able to use the ve-

getation classes as a means to interpolate the plant diversity indices to the entire

study area. If a strong association were found, the mapped classes would serve as a

mechanism for up-scaling measurements of plant diversity on the ground to the

entire landscape. This process can be considered as a typical case of global in-

terpolation of point-data (Borrough and McDonnell 1998).

To estimate the biological composition of species at any location in the entire

studied area, the average plant diversity was calculated for every one of the ve-

getation classes. The problem is reduced to finding out how useful the mean di-

versity values of vegetation classes are, to predict the plant diversity values at

unvisited sites. The approach essentially makes use of the classes for interpolating

their mean values of plant diversity to the entire area covered by these classes

2607

(Ponce-Hernandez 1987; Leenhardt et al. 1994; Borrough and McDonnell 1998).

This method also assumes that dissimilates of diversity values within vegetation

classes are significantly lower than those between them. That is to say, the

within-class variability of the diversity values needs to be confirmed as sub-

stantially smaller (i.e., internal homogeneity) than that between classes, before a

class can be used for this interpolation. The model to predict plant diversity values

using a classified map is:

Zij ¼ �þ vtj þ "ij

where Zij is the value of the plant diversity index at sampling quadrat i within the

vegetation type j. The parameter m is the general mean of the plant diversity index,

vtj is the difference between the m and the mean of class j and eij is the random error.

The estimate of Zij is given by the mean value of the observations within the class j.

A standard analysis of variance was used to test for significant differences in

mean plant diversity composition across vegetation classes. Additionally, to provide

an indicator of the degree of internal uniformity of each of the vegetation

classes with respect to plant diversity values, a uniformity index was constructed.

This index could be used as a measure of the goodness of the classification in

partitioning spatial variability and it is analogous to r2 in regression analysis

(Leenhardt et al. 1994). The uniformity index is composed by subtracting from 1

the relative variance, which is a ratio of the within-class variance and total sample

variance. The equation for calculating the uniformity index (U) is:

U ¼ 1 � �2W

�2T

When the value of the index is close to 1, the mapped classes are very uniform and

meaningful in terms of partitioning the variability of plant diversity indices across

the landscape. This can also be viewed as the existence of a strong relationship

between the spatial distribution of plant diversity indices and the mapped classes or

as the goodness of prediction of the mean values of each vegetation class. There-

fore, the classes might be useful for discriminating plant species composition of

each vegetation type.

Results

Land cover map output

The land cover raster map of the studied area resulting from the supervised clas-

sification is shown in Figure 3. The total area occupied by this landscape is

6464.2 ha, of which 84.98% are covered by tropical sub-deciduous forest including

all successional stages (kanah kax, kelenche, juche, saakab) and by two secondary

associations (akalche, savanna). Deforested areas, grassland, crops and water

bodies comprise the other 15.02%. Kanah kax is the largest land cover type

stretching over 36.86% of the total area with 2382.75 ha. Kelenche has 26.27% of

2608

the total area with 1698.03 ha. Juche occupies 11.93% of the total area with

771.66 ha whereas akalche and savanna have an area that covers 9.73% of the total

area.

The ability to use the vegetation map, derived from classified multispectral sa-

tellite imagery, as a good predictor and as a reliable mechanism for up-scaling

ground plant diversity estimates, depends to a large extent on the accuracy with

which the identified classes in the image depict the true vegetation classes on the

ground. The results of accuracy assessment are presented in Table 1. The overall

accuracy of the map was calculated to be 82.3, while the kappa index was 0.78. The

greatest misclassifications among the vegetation-mapped classes occurred in the

kelenche class (relative accuracy of this vegetation type in the produced map is

76.0%). This can be explained by the fact that the ‘kelenche’ class is not well

differentiated in terms of development and species composition from its neigh-

boring classes: Kanah kax and juche. The boundaries between these three classes

are fairly gradual and transitional, making the classes more inclusive and therefore

with greater internal variability.

Documenting completeness of the sampling

The fitted models of the species accumulation curves were able to evaluate the

completeness of plant species inventories. Both LD and Clench models were useful

predictors of the number of species (r2> 0.9, Table 2). The LD model, however,

Figure 3. Land cover map of the study area showing the six vegetation types mapped from a supervised

classification.

2609

reached the asymptote with a lower number of species than the Clench model for

every one of the vegetation types (Figure 4) and for the landscape (Figure 5).

The number of species observed (recorded on the ground), in all the vegetation

types, is larger than that predicted by the LD model, except for the Saakab class that

has 92 species compared with 94 predicted by the LD model (Table 3), which

represents 98.2% of the asymptote. This indicates that according to this model, the

number of sample sites was more than adequate in obtaining a fair representation of

the number of species in the studied area. In contrast, according to the values

computed with the Clench model none of the vegetation types reaches the

asymptote, but at least all the classes have as a minimum 83.3% of the total species

predicted by this model. Again, the exception was the ‘Saakab’ class, which has

only 74.2% of the species predicted. This is probably due to the fact that Saakab

includes species between 2 and 4 years of age, as well as some trees. Thus, most of

the information for counting the diversity of this class comes from a sub-quadrat of

5 m� 5 m. Consequently, there is less total area sampled in this class than the area

sampled for the rest of the classes.

Table 1. Error matrix for the land cover map of the study area.

Classes in Quadrats sampled at field User’s

the imageKanah kax Kelenche Juche Saakab Akalche Savanna Total

accuracy

Kanah kax 35 2 37 94.6

Kelenche 4 19 2 1 2 28 67.9

Juche 2 2 17 1 22 77.3

Saakab 1 17 2 20 85.0

Akalche 1 14 1 16 87.5

Savanna 1 2 14 17 82.4

Grass and crops 1 1

Total 42 25 20 20 17 17 141

Producer’s accuracy 83.3 76.0 85.0 85.0 82.4 82.4 Accuracy¼ 82.3;

Kappa¼ 0.78

Table 2. Regression statistics for the two species accumulation models fitted for each vegetation type

and for the total landscape.

Vegetation type Linear dependence model Clench model

S¼ a=b(1�exp(�b*sample)) S¼ a*sample =(1þ b*sample)

a b r2 a b r2

Kanah Kax 19.68 0.16 0.91 30.32 0.21 0.98

Kelenche 20.77 0.19 0.94 28.95 0.21 0.98

Juche 22.20 0.21 0.98 29.63 0.23 0.99

Saakab 14.37 0.15 1.00 17.26 0.14 1.00

Akalche 4.69 0.24 0.97 6.22 0.26 0.99

Savanna 2.57 0.39 0.91 3.94 0.52 0.98

Landscape 9.07 0.05 0.90 14.42 0.07 0.98

2610

The number of quadrats needed to achieve 90 and 95% of the total species

estimated by the Clench model is shown in Table 3. These values represent the

sample effort. For example, to increase the number of species recorded in Kelenche

from 117 (86.8% of the total species predicted by the Clench Model) to 121 species

Figure 4. Species accumulation curves of number of plant species for six vegetation types. The curves

were created using observed data and estimates of fitted models.

2611

(90% of total predicted by the same model), it would be required to increase the

sample size from 25 to 42 quadrats. That is equivalent to employing almost twice

the initial sample effort, to increase the number of species in the kelenche class by

four. The inclusion of one extra species within this class becomes increasingly

expensive in terms of sampling effort. Considering that the survey sampling to

estimate plant species in the study area has achieved over 98% of the number of

species in one of the models (LD) and at least 83% in the other (Clench Model),

these values are considered satisfactory and the inventories of plants species in the

different communities (vegetation types) are comparable.

Figure 5. Species accumulation curves of number of plant species for entire landscape. The curves

were created using observed data and estimates of fitted models.

Table 3. Comparison of number of species and quadrats among observed data and predictions of two

species accumulation models for each vegetation type and the landscape.

Vegetation Observed LD model Clench model

typeNo. of

spp.

No. of

quadrats

No. of

spp.

% of the

asymptote

No. of

spp.

% of the

asymptote

No. of

quadrats*

90% 95%

Kanah kax 133 42 124 107.4 143 93.3 42 89

Kelenche 117 25 112 104.9 135 86.8 42 88

Juche 110 20 106 104.2 130 84.4 40 84

Saakab 92 20 94 98.2 124 74.2 65 137

Akalche 20 17 19 103.1 24 83.3 35 73

Savanna 7 17 7 105.8 8 91.9 17 37

Landscape 182 141 169 107.7 203 89.9 126 267

*Number of quadrats required to reach 90 and 95% of asymptote.

2612

Plant species diversity

A total of 16543 individuals were sampled in 141 plot sites (quadrats). In this

sample 182 plant species were recorded, which belong to 53 family groups. Of the

total number of species identified 62.4% are trees, 14.0% are shrubs and 23.6% are

vines, palms and some herbs.

Forest sub-deciduous vegetation types (Kanakax, kelenche, juche, and saakab)

have the highest values of species richness in the landscape, from 92 to 133 species.

In contrast, the number of species in the secondary associations (Akalche and

Savanna) varies from 7 to 20 species (Table 4). The forest classes also have fewer

abundant species and more rare species than secondary associations. This indicates

that these classes are not dominated by a few species, which is shown in the high

Table 4. Trees, shrubs and vines diversity measures for each vegetation type and the entire landscape by

height class.

Vegetation

type

Layer (height class) No. of

species

No. of

individuals

Shannon

(H)

Evenness

(H=ln S)

Simpson

(1�D)

Kanah Kax > 20 m 11 20 2.27 0.95 0.93

> 10 m and 20 m 55 322 3.27 0.82 0.94

> 3 m and 10 m 127 4322 3.91 0.81 0.97

> 1 m and 3 m 71 837 3.35 0.78 0.93

Total 133 5501 3.92 0.80 0.97

Kelenche > 10 m and 20 m 30 114 2.91 0.86 0.93

> 3 m and 10 m 114 3551 3.67 0.77 0.96

> 1 m and 3 m 62 310 3.47 0.84 0.95

Total 117 3975 3.70 0.78 0.96

Juche > 3 m and 10 m 103 3224 3.58 0.77 0.95

> 1 m and 3 m 62 366 3.66 0.89 0.97

Total 110 3590 3.68 0.78 0.96

Saakab > 3 m and 10 m 35 118 3.11 0.87 0.96

> 1 m and 3 m 91 1270 3.58 0.79 0.94

Total 92 1388 3.68 0.81 0.95

Akalche > 10 m and 20 m 2 16 0.38 0.54 0.23

> 3 m and 10 m 18 1394 0.89 0.31 0.38

> 1 m and 3 m 13 138 1.61 0.63 0.65

Total 20 1548 1.00 0.33 0.41

Savanna > 3 m and 10 m 6 230 1.17 0.65 0.59

> 1 m and 3 m 5 311 0.39 0.24 0.38

Total 7 541 1.09 0.56 0.55

Landscape 182 16543 4.06 0.78 0.97

2613

values of Simpson and Shannon indices. The evenness in these classes is also

greater than those of secondary associations; these are dominated by two species,

H. campechianum in akalche and S. pterota in savanna, as they show the lowest

values of Simpson and Shannon indices. The plant diversity among forest classes,

as intuitively expected, has the highest values in the climax stage (kanah kax) and the

lowest values were found in the first stage of the succession (saakab). For example,

in kanah kax 133 species were recorded while in kelenche and juche 117 and 110

species were found. Finally, in saakab only 92 species were identified.

Most of the diversity in the six vegetation types comes from the layer between 3

and 10 m of height, except for saakab. For example, of the total 133 species found

in kanah kax, 127 were recorded in the layer between 3 and 10 m of height. The

diversity in saakab is mostly explained by a layer between 1 and 3 m with 91 of a

total of 92 species found in the class. This is due to saakab having the composition

of its vegetation made up by plants from 2 to 4 years of age.

Another way of comparing the diversity among the six communities is using the

average value of the sampling quadrats, in other words, the mean value of number

of species, Shannon and Simpson indices in an area of 100 m2. The mean values of

diversity indices for the six vegetation types showed a similar, if not the same

pattern as described above for the total a-diversity. Thus, forest classes have more

species and less dominance than secondary associations (P< 0.001). Further, the

kanah kax class (i.e. the mature forest) has more species and less dominance than

kelenche, juche and saakab (P< 0.001), which are early successional stages of the

forest (Table 5).

The rate of species turnover among forest classes measured by Whittaker’s index

ranged from 0.17 to 0.40. The Inverse of Jaccard measure produces similar results

to those of Whittaker’s index for the forest classes. Their values vary from 0.29 to

0.57. However, the change in species composition among forest classes and sec-

ondary associations was noticeably larger than those among forest classes. Their

values ranged from 0.77 to 0.99 (Table 6). This indicates that although the sec-

ondary associations have low values of alpha diversity, they have an important

contribution to the total diversity of the landscape.

Relating the mapped vegetation classes to biodiversity indices

There were significant differences among the mean a-diversity values for number of

species, exp Shannon and reciprocal Simpson indices of the six vegetation types

(p< 0.00001, Table 7). In terms of the uniformity of mapped classes, as reflected by

the three diversity indices mentioned above, the results also confirm that the ve-

getation classes had a relatively high uniformity in two of the plant diversity indices

(number of species and exp Shannon index), and a moderate uniformity value for

the reciprocal Simpson index. This shows that the vegetation map generated from

the supervised classification is useful in partitioning the spatial variability of the

three diversity indices over the study area, and therefore it can be used for the

purpose of mapping and interpolating plant diversity.

2614

Ta

ble

5.

Mea

nal

pha

div

ersi

tyval

ues

of

tree

s,sh

rub

san

dv

ines

and

thei

rst

and

ard

dev

iati

on

insi

xveg

etat

ion

typ

es.

An

LS

Dte

stw

asp

erfo

rmed

toco

mp

are

the

mea

n

div

ersi

tyval

ues

among

the

veg

etat

ion

types

.

Veg

etat

ion

type

No.

of

spec

ies

Exp

Shan

non

Rec

ipro

cal

Sim

pso

n

Fie

ldq

uad

rats

**

Map

qu

adra

ts*

**

Fie

ldq

uad

rats

Map

qu

adra

tsF

ield

qu

adra

tsM

apq

uad

rats

Kan

aK

ax3

4.9

1�

5.3

83

5.1

1�

5.6

02

1.9

1�

5.5

72

2.0

2�

5.8

21

5.2

0�

5.4

11

5.2

9�

5.6

4

Kel

ench

e31.9

2�

4.9

13

1.7

4�

4.6

21

8.8

8�

3.5

11

9.0

3�

3.5

61

2.9

7�

3.1

51

3.1

8�

3.4

7

Juch

e2

8.3

5�

4.2

72

8.7

1�

4.5

41

5.3

0�

4.3

21

5.3

7�

4.5

41

0.0

4�

3.6

81

0.0

0�

3.8

4

Saa

kab

15

.80�

4.5

41

6.2

4�

4.6

99

.26�

4.2

49

.26�

4.5

46

.68�

3.7

36

.54�

3.9

7

Ak

alch

e5

.71�

1.7

25

.86�

1.8

32

.30�

0.6

4*

2.4

1�

0.6

4*

1.7

1�

0.3

8*

1.7

8�

0.3

7*

Sav

ann

a3

.00�

0.7

93

.14�

0.7

72

.32�

0.4

1*

2.4

1�

0.4

0*

2.0

5�

0.2

9*

2.1

2�

0.2

8*

*N

osi

gn

ifica

nt

dif

fere

nce

sb

etw

een

thes

eg

rou

ps.

**

Sam

ple

dq

uad

rats

inth

efi

eld

.

**

*S

amp

led

qu

adra

tsco

rrec

tly

clas

sifi

edin

the

map

.

2615

On the other hand, comparing the performance in terms of the accuracy of pre-

dictions, for the sampled units in the field and for those samples that were correctly

classified in the map (Table 7), the uniformity indices of the three diversity values

were very similar. For example, the number of species shows a uniformity index of

0.89 for the quadrats sampled in the ground, while the quadrats that were well

classified in the vegetation map have a uniformity index of 0.88. This indicates that

the samples in the classified map are as accurate in predicting mean plant diversity

values as the means strictly derived from ground observations. Therefore, the per-

formance of both data sets for the purpose of prediction gave very similar results.

Table 6. Whittaker beta diversity values and Inverse Jaccard similarity

measure between pairs of vegetation types in the landscape.

Vegetation type Whittaker’s beta diversity Inverse of Jaccard measure

Kanah kax

Kelenche 0.17 0.29

Juche 0.25 0.40

Saakab 0.40 0.57

Akalche 0.80 0.89

Savanna 0.97 0.99

Kelenche

Juche 0.23 0.38

Saakab 0.39 0.56

Alakche 0.78 0.88

Savanna 0.97 0.98

Juche

Saakab 0.33 0.49

Alakche 0.77 0.87

Savanna 0.97 0.98

Saakab

Alakche 0.77 0.87

Savanna 0.96 0.98

Akalche

Savanna 0.85 0.92

Table 7. One factor between subjects ANOVA of several plant diversity measures testing for difference

among vegetation types.

Dependent variable DF F Uniformity index

Quadrats sampled at field

Number of species [5,140] 225.28* 0.89

Exp Shannon [5,140] 99.10* 0.78

Reciprocal Simpson [5,140] 52.17* 0.65

Quadrats correctly classified in the map

Number of species [5,115] 176.44* 0.88

Exp Shannon [5,115] 74.12* 0.76

Reciprocal Simpson [5,115] 39.08* 0.62

*P< 0.00000.

2616

Least significant difference tests were applied to ascertain whether the plant

composition, measured by the three diversity indices (number of species, exp

Shannon and reciprocal Simpson), was significantly different among all vegetation

classes. The six vegetation types are significantly different in all: number of spe-

cies, exponent Shannon and reciprocal Simpson (P< 0.001), with the exception of

the akalche and savanna classes, which cannot be discriminated by two diversity

indices (i.e., exp Shannon and reciprocal Simpson) (P> 0.05, Table 5). This may be

due to the high degree of similarities in structure and composition of these two

vegetation types, hence the difficulty in separating them. Even though both sets of

data (field quadrats and map quadrats) have similar mean values for each of the six

vegetation types in the three diversity indices, the results of the map quadrats show

slightly larger variances (Table 5).

Discussion

The general objective of this paper was to develop an approach for mapping plant

species diversity in a tropical forest. However, biodiversity mapping, to be useful

and to obtain credible results, must consider a number of important technical issues;

among them are the accuracy of the map (Williams 1996), the relationship among

the mapped classes and the plant diversity measures (Borrough and McDonnell

1998), the power of representation of the sample species data (Soberon and

Llorente 1993; Moreno and Halffter 2002), and the need for considering species

turnover as a component of biodiversity (Whittaker 1972).

The studied area is largely forested (about 85% of the total area). The forest is in

a variety of stages of succession, from 2 to 3 years fallow after slash-and-burn

agriculture, to up to 50 of more years of a mature forest. Such variations in the

stages of succession create transitional changes, which should translate into tran-

sitional boundaries rather than sharp or ‘crisp’ ones. Naturally occurring sharp

boundaries after being mapped out would have avoided inclusions or ‘impurities’

into classes. The fact that many of these boundaries may be transitional in nature,

contributes to the presence of inclusions. Hence, the success in identifying and

mapping relatively homogeneous vegetation classes was strongly influenced by the

degree of differentiation of the successional stages of the forested mass (Table 1).

This calls into question whether a crisp mapping model is suitable for re-

presenting classifications of these tropical forests (Borrough and McDonnell 1998).

The overall accuracy of the map was calculated to be 82.3%. These figures show a

reasonably accurate map of vegetation classes, despite of the factors discussed

above. Similar findings have been reported elsewhere in the world. For instance,

Fuller et al. (1997) reported an accuracy map of 86% whereas Nagendra and Gadgil

(1999) reported 88% of accuracy in their classification.

The relationship that exists between the structure of the landscape and species

diversity can be used to design mechanisms for the assessment of biodiversity at the

landscape scale. Remote sensing provides a useful spatial framework for the col-

lection and analysis of data, and for exploring these relationships, which can be

2617

used in predicting and up-scaling site measurements of biodiversity to the land-

scape scale (Stoms and Estes 1993; Innes and Koch 1998). Thus, classification of

the landscape from satellite imagery into relatively homogeneous units (habitat

types) can be used for estimating the status of biodiversity if it can be proved that a

relatively strong degree of association exists between the boundaries of classes and

the spatial variability of the biodiversity indices.

This study has shown that vegetation classes identified in the field and mapped

through satellite imagery were sufficiently associated with the spatial variability of

plant diversity indices, to allow the former to be used for the stratification of the

latter. The stratification of plant composition as measured by the three diversity

indices, into the mapped vegetation classes, was statistically significant among all

vegetation classes. There were, however, some exceptions, particularly between the

akalche and savanna classes. Even though the number of species can differentiate

them, both share a similar structure in the abundance of species.

To improve the accuracy of predictions for the three diversity indices under the

approach presented in this paper, the within-class variance should be further re-

duced. This can be done by either achieving a more accurate classification of the

FCC image, or by subdividing the vegetation types more finely so that finer

splinters in the current classes can be recognized and their number increased.

However, it is difficult to see how the satellite image classification process could

discriminate habitats at finer scales without changing the image fixed resolution

(Turner et al. 2003). Alternatively the within-class variance could be reduced by

possibly increasing the sample size. This has its very practical implications in terms

of greater cost, time and effort. There are trade-offs between achieving greater

accuracy of prediction by classes through further field sampling, and the costs

incurred in it. At some point, the conceptual trade-off curve becomes asymptotic,

with little gains in accuracy. The results presented here for samples sites in the

ground and those quadrats correctly classified in the image are very near that point.

Tables 7 and 5 show similar results in accuracy of prediction and values of diversity

with two sample sizes. However, there should be a practical threshold of sampling

to gain accuracy of predictions.

One of the main concerns when the biodiversity of a landscape is mapped, and

the diversity of the different habitats is compared, is to obtain an adequate and

representative sample. Usually there is no direct measure of the degree of re-

presentation achieved by the sample size, but this can be indicated by species

accumulation curves (Soberon and Llorente 1993) and other non-parametric

methods (Colwell and Coddington 1995). In this study, the inventories of plants

demonstrate a satisfactory level of completeness. Our plant diversity estimates were

over the LD model and the number species estimated in the sample reached at least

80% of the other estimates (Table 3). There was however, an exception; the saakab

class had 74% of the estimated value by the Clench model, but with 1388 in-

dividuals sampled. Comparing these results with other studies, Slocomb and

Dickson (1978), cited by Baltanas (1992), suggested that the sample size could be a

problem when the sample does not have at least 80% of the total number of species

or it does not have at least 1000 individuals sampled.

2618

Diversity in the study area shows that forest classes have higher a-diversity

values than secondary associations. However, the overall diversity of the landscape

is not completely due to the forest areas because akalche and savanna possess

unique species that contribute to the total biodiversity of the landscape mosaic. This

can be appreciated through the high b-diversity values among forest types and

secondary associations, as a result of the different habitat conditions in which they

are growing (e.g. secondary associations are found in flooded areas, see Table 6).

Examining diversity among forest classes reveals that plant diversity increases

significantly during succession (Table 4), showing consistent results with other

studies (Turner et al. 1997). Nevertheless, the floristic and species composition

found in a tropical forest after disturbance could have a notable variation among

regions, giving as a result different successional systems (Finegan 1996). These

variations can be attributable to the way in which the physical and biotic factors

change during this period of time (Gomez-Pompa and Vazquez 1983).

There is a difference in species composition among mature forests (kanah kax,

kelenche) and those of early stages of succession (saakab), represented by moderate

b-diversity values, which are around 40% (Table 6). This implies that the total

diversity of the forest is not exclusively the result of the areas with mature forest.

The species turnover among the different stages of succession is due to the live

history traits of the species in them (Gomez-Pompa 1971). In the first phase of the

succession, the forest is occupied by species that have continuous production of a

large number of small and widely dispersed seeds. They also possess relatively high

photosynthetic capacities and low investment of resources in stems and branches,

but with fast growth in diameter and height. These species are replaced during

succession by others, which have larger seeds of shorter viability, with lower

photosynthetic capacity, which confer slower growth and longer lifespan (Finegan

1996). This successional change is attributed to competition among individuals of

those species and their tolerance to microclimatic changes (Gomez-Pompa and

Vazquez 1983).

Conclusions

Our results lead us to the following conclusions. First, the identification of vege-

tation classes on the field and their mapping using satellite image classification

were found to discriminate and separate significantly different species composi-

tions, in such a way that they can provide a useful mechanism for interpolating, and

up-scaling values of diversity indices over the entire landscape at unvisited loca-

tions within a given class. Second, to estimate and map out plant species diversity

over the area special attention needs to be placed on assessing the power of re-

presentation of the sample sites and the incorporation of beta diversity as a com-

ponent of plant diversity.

The results also suggest the existence of a pattern in terms of plant diversity and

vegetation classes representing different stages of forest succession. The older the

stage of forest succession, the more species it contains and the less dominance there

2619

is of any given species. However, the total plant diversity in the landscape is com-

plemented by the species found in the early stages of succession and in the secondary

associations found in the studied area. For the purpose of conservation of plant

diversity in the area, these findings support not only the intuitive notion of protecting

mature forests, the most diverse community in the area, but also to preserve the

diversity of young successional forests and that of secondary associations.

References

Baltanas A. 1992. On the use of some methods for estimation of species richness. Oikos 65:

484–492.

Borrough P.A. and McDonnell R.A. 1998. Principles of Geographical Information Systems. Spatial

Information Systems and Geostatistics. Oxford University Press, Oxford, UK.

Cabrera C.E., Souza S.M. and Tellez V.O. 1982. Imagenes de la flora Quintanaroense. Centro de

Investigaciones de Quintana Roo, Mexico.

Campbell J.B. 1987. Introduction to Remote Sensing. The Guilford Press, New York.

Carroll S.S. 1998. Modeling abiotic indicators when obtaining spatial predictions of species richness.

Environmental and Ecological Statistics 5: 257–276.

Colwell R.K. 1997. EstimateS: statistical estimation of species richness and shared species from samples,

version 5.0.1. User’s guide and application (htpp://viceroy.eeb.uconn.edu/estimates).

Colwell R.K. and Coddington J.A. 1995. Estimating terrestrial biodiversity through extrapolation. In:

Hawksworth D.L. (ed) 1995. Biodiversity Measurement and Estimation. Royal Society: Chapman &

Hall, London.

Duch J.G. 1995. Los suelos, la agricultura y vegetation en Yucatan. In: Hernandez X.E., Bello E. and

Levy S. (eds) La milpa en Yucatan. Colegio de Postgraduados, Mexico, pp. 97–108.

Earth Resource Mapping Ltd. 1998. ER mapper 6.1. User guide. Earth Resource Mapping, San Diego,

California.

Finegan B. 1996. Pattern and process in neotropical secondary rain forest: the first 100 years of suc-

cession. Trends in Ecology and Evolution 11: 119–124.

Forman R.T.T. 1995. Land Mosaics, the Ecology of Landscapes and Regions. Cambridge University

Press, Cambridge, UK.

Fuller R.M., Groom G.B., Mugisha S., Ipulet P., Pomeroy D., Katende A. et al. 1997. The integration of

field survey and remote sensing for biodiversity assessment: a case study in the tropical forest and

wetlands of Sango Bay, Uganda. Biological Conservation 86: 379–391.

Gomez-Pompa A. 1971. Posible papel de la vegetacion secundaria en la evolucion de la flora tropical.

Biotropica 3: 125–135.

Gomez-Pompa A. and Vazquez-Yanez C. 1983. Estudio sobre la sucesion secundaria en los tropicos

calido-humedos: el ciclo de vida de las especies secundarias. In: Gomez-Pompa A., del Amo S.,

Vazquez-Yanez C. and Butanda A. (eds) Regeneracion de selvas. INIREB, Mexico, pp. 599–613.

Groombridge B. 1992. Global Biodiversity. Status of the Earth’s Living Resources. Chapman & Hall,

London.

Innes J.L. and Koch B. 1998. Forest biodiversity and its assessment by remote sensing. Global Ecology

and Biogeography Letters 7: 397–419.

Isik K., Yaltikik F. and Akesen A. 1997. The interrelationship of forests, biological diversity and the

maintenance of natural resources. Unasylva FAO 48: 190–191.

Kerr J.T. and Ostrovsky M. 2003. From space to species: ecological applications for remote sensing.

Trends in Ecology and Evolution 18: 299–305.

Krebs C.J. 1989. Ecological Methodology. Harper Collins, New York.

Lamas G., Robbins R.K. and Harvey D.J. 1991. A preliminary survey of the butterfly fauna of Pakitza,

Parque Nacional del Manu, Peru, with an estimate of its species richness. Publicaciones del Museo de

Historia Natural, Universidad Nacional Mayor de San Marco, A40: 1–19.

2620

Leenhardt D., Voltz M., Bornand M., and Webster R. 1994. Evaluating soil maps for prediction of soil

water properties. European Journal of Soil Science 45: 293–301.

Magurran A.E. 1988. Ecological Diversity and Its Measurement. Princeton University Press, Princeton,

New Jersey.

Mazerolle M.J. and Villard M.A. 1999. Patch characteristics and landscape context as predictor of

species presence and abundance: a review. Ecoscience 6: 177–124.

Miller R.I. and Wiegert R.G. 1989. Documenting completeness, species–area relations, and the species

abundance distribution of a regional flora. Ecology 70: 16–22.

Moreno C.E. and Halffter G. 2000. Assessing the completeness of bat biodiversity inventories using

species accumulation curves. Journal of Applied Ecology 35: 13–23.

Moreno C.E. and Halffter G. 2001. Spatial and temporal analysis of a, b, and g diversities of bats in a

fragmented landscape. Biodiversity and Conservation 10: 367–382.

Muldavin E.H., Neville P. and Harpper G. 2001. Indices of grassland biodiversity in the Chihuahuan

desert ecoregion derived from remote sensing. Conservation Biology 15: 844–855.

Myers N., Mittermeier R.A., Mittermeier C.G., da Fonseca G.A.B. and Kent J. 2000. Biodiversity

hotspots for conservation priorities. Nature 43: 853–858.

Nagendra H. and Gadgil M. 1999. Satellite imagery as a tool for monitoring species diversity: an

assessment. Journal of Applied Ecology 36: 388–397.

Ponce-Hernandez 1987. The use of soil information systems for land planning: procedures for gen-

erating information for properties, depths and sites not recorded. D.Phil. Thesis, University of Oxford,

Oxford, UK.

Sarukhan J. and Pennington T.D. 1998. Arboles tropicales de Mexico. Universidad Autonoma de Mexico

Fondo de Cultura Economica, Mexico.

SEMARNAP 1998. Los recursos forestales de Mexico. Subsecretaria de Recursos Naturales, Direccion

General Forestal, Mexico.

Soberon J.M. and Llorente J.B. 1993. The use of species accumulation functions for the prediction of

species richness. Conservation Biology 7: 480–488.

Sosa V., Flores J.S., Rico-Gray V., Lira R. and Ortiz J.J. 1985. Etnoflora Yucatense. Lista floristica y

sinonimia Maya. Instituto Nacional de Investigaciones sobre Recursos Bioticos, Veracruz, Mexico.

Stoms D.M. and Estes J.E. 1993. A remote sensing research agenda for mapping and monitoring

biodiversity. International Journal of Remote Sensing 14: 1839–1860.

Turner I.M., Wong Y.K., Chew P.T. and Ibrahim Ali bin 1997. Tree species richness in primary and old

tropical forest in Singapore. Biodiversity and Conservation 6: 537–543.

Turner W., Spector S., Gardiner N., Fladeland M., Sterling E. and Steininger M. 2003. Remote sensing

for biodiversity science and conservation. Trends in Ecology and Evolution 18: 306–314.

Whigham D.F., Olmsted I., Cabrera Cano E. and Harmon M.E. 1991. The impact of hurricane Gilbert on

trees, literfall and woody debris in a dry tropical forest in the northeastern Yucatan peninsula. Bio-

tropica 23: 434–441.

Whittaker R.A. 1972. Evolution and measures of species diversity. Taxon 21: 213–251.

Williams B.K. 1996. Assessment of accuracy in the mapping of vertebrate biodiversity. Environmental

Management 47: 269–282.

2621