Bennett C. Sandler, Peter B. Pearman, Mauricio Guerrero G., and Karen Levy.

Using a GIS to Assess Spatial Scale of Taxonomic Richness in Amazonian Ecuador

ABSTRACT

Multi-taxonomic sampling is a common tool for quantifying effects of disturbance on biological diversity. Results and subsequent management implications are significantly affected by the spatial configuration of the sampling design employed. To assess disturbance effects, sampling should control for variability inherent in species counts as well as variation due to natural environmental gradients. The Center for Conservation Biology at Stanford University has developed a methodological approach using GIS to determine a scale-sensitive sampling scheme for measuring changes in taxonomic diversity along a disturbance gradient. Sample data were collected at Jatun Sacha Biological Station in Napo Province, Ecuador where primary tropical forest conversion is advancing at an alarming rate due to human colonization. Workstation ArcInfo was used to extract and model a variety of landscape variables, including "imbeddedness" in primary forest, land cover diversity, elevation, and solar insolation, at varying neighborhood sizes around sample sites, to test their value for predicting species richness. The GRID module in ArcInfo was used to extract a series of landscape metrics at varying distances from sites where intensive species surveys were conducted. In addition, A/I was used to generate a digital elevation model from which many of the landscape variables were derived. PCI image processing software was used to classify land cover from Landsat Thematic Mapper data. Pfinder software was used to differentially correct GPS data. SAS and JMP were used for statistical tests. Results from this research offer explicit spatial decision-rules for designing monitoring studies to test predicted effects of human disturbance on taxonomic diversity.



Introduction

The current human-induced trend of increasing species extinction (Wilson 1988) presents a mandate to conservation biologists to elucidate natural patterns of species distribution and assemblages and the influence of perturbation on these patterns. Patterns indicate processes at work which order ecological communities (Allen and Starr 1982, Noss 1990). The critical, and often difficult task, of finding order in species distributions and assemblages is identifying the spatial scale at which those patterns occur (ESA 1991).

Community dynamics are further complicated by a lack of understanding of how different taxonomic groups, which may exhibit distribution patterns at different scales, covary in their response to disturbance (Pearson and Cassola 1992, Prendergrast et al. 1993, Gaston 1996). Ideally, resource managers would like to monitor a select set of indicator species with a minimum of sampling effort as a means for evaluating overall species richness or the impact of different management regimes (Noss 1990, Gaston 1996).

Geographers and geo-statisticians offer conservation biologists a suite of analytical tools to help assess scale and design effective inventory and monitoring schemes that capture important distribution patterns. These tools include variogram analysis (Davis et al. 1991, Isaaks and Srivastava 1989) and spatial autocorrelation indices (Isaaks and Srivastava 1989, Webster and Oliver 1990). As these tools become more commonly used in ecological research, we should gain a greater understanding of both the environmental constraints and intra- and inter-specific interactions that shape biotic assemblages.

This paper describes methods used to qualitatively and quantitatively assess spatial distribution patterns of bird, butterfly and understory bird species richness in Amazonian Ecuador and correlate richness to underlying environmental gradients.

Study Area

Taxonomic surveys were conducted at Jatun Sacha Biological Station and Reserve in the Napo Province of Ecuador. The 3600 ha reserve is situated 30 km east of the Andean Cordillera at approximately 450 meters above sea level (7736íW, 104íS) and abuts the Napo River. Natural habitats range from floodplain forest and Heliconia swamp to terra firma forest on steeply dissected ridges and swales. Approximately 70 percent of the reserve is covered in primary forest, the balance consisting of selectively logged forest, clear-cut regrowth of varying age, or abandoned orchard. Species richness within the reserve is high including over 1100 species of trees, 500 species of birds and 69 species of frogs (Pearman et al, 1995).

Sampling Design

Between January 1994 and February 1995, understory birds, amphibians, and butterflies were surveyed at four different times at all 23 sites throughout the reserve. Sites were located in either pristine forest, selectively logged forest, or abandoned orchard at varying distance from pasture (between 15 and 1300 meters). Sites were minimally located 150 meters away from each other.

Different sampling methods were used for each taxonomic group. Birds were captured using mist nets and amphibians were sampled using visual encounter surveys. Over the duration of the study period, each site was randomly sampled twice for birds (220 mist net hours /site) and four times for amphibians. Butterflies were sampled once in January, February, and September using baited tube traps (see Pearman et al., submitted) and twice between August and September using sweep nets (Levy 1995). Sampling effort was equal among sites. Sampling order was randomized across sites within each sampling period.

Sites were georeferenced using a Trimble GeoExplorer global positioning system, differentially corrected to 5 meters. Site centroids were calculated from geo-referenced transects within each site.

Input Spatial Data

Two classes of landscape, or environmental, variables were generated for input in this study. Physiographic variables, including elevation, slope, aspect, and insolation were all derived from a digital elevation model (DEM) of the area. Land cover statistics such as proportion of primary forest and land cover diversity index were derived from a supervised classification of Landsat Thematic Mapper (TM) imagery.

Digital Elevation Model

The digital elevation model was generated from twenty meter contours and hydrologic features digitized off of the Puerto MisahuallÌ map sheet (Instituto Geogr·fico Militar, 1988) of the Ecuadorian national 1:50,000 base map series. Contour and hydrologic vectors from the reserve area were manually digitized using a Calcomp 9500 tablet attached to a Sun SPARCII workstation running ArcInfo 7.0.4. Error associated with the transformation of vectors from digitizer units to UTM projection coordinates (Zone 18, South America 1956) was 5.428 meters (rmse). The DEM was then constructed using ArcInfoís TOPOGRID command, using both the contour data and hydrology. Cell size was set to 28.5 meters to match the resolution of the Landsat TM imagery. The DEM was visually assessed for errors by inspecting a rendered hillshade surface as well as by inspecting output sink and drain diagnostic coverages. Several attributes of the input hydrology vectors required editing before an acceptable DEM was completed.

Slope and aspect surfaces were easily calculated in GRID from the completed DEM. Solarflux, an ArcInfo AML program written (Hetrick et al. 1993) was run to generate a surface model of incoming solar insolation for June 21. Insolation is a more direct measure of temperature variation than using slope and/or aspect as a surrogate.

Land Cover

We were unable to find cloud free TM imagery in the archives of EOSAT for any time during the two years field work was being conducted (1994-1995) so a 100 kilometer by 100 kilometer subscene of Path 9, Row 60 from July 14 1992 was purchased from EOSAT. A smaller window of this image was extracted and geocoded to the UTM projection (Zone 18, South America 1956 Datum) using eight ground control points selected within the Puerto MisuhuallÌ map (1:50,000). Error (rmse) associated with rubbersheeting the imagery was less than one half pixel (approximately 15 meters). All image processing was performed using PCI image processing software on a Silicon Graphics Indigo II Extreme workstation.

Once the imagery was geo-coded, a supervised classification was performed using a very simple information class scheme: primary forest, secondary forest, agriculture, bare ground/sand, and water. Classes were trained visually with the assistance of Pearman, who lived at Jatun Sacha for two years while supervising the field research. Classification accuracy was also evaluated through visual inspection. Spectral separation of information classes was robust except for slight confusion between water and primary forest biased by topographic shading, and between primary forest along high reflectance ridge tops that were misclassified as secondary forest. Classification errors were not thought to be significant enough to warrant systematic post-classification processing. Only one of the twenty three sample sites was misclassified. The land cover assignment at this site (represented by four pixels) was manually reclassified from secondary forest to primary forest to ensure accuracy.

Analysis

Once the survey data, DEM and derived coverages, and land cover classification were entered into the GIS and checked for accuracy, spatial analysis was initiated. The scope of this analysis was limited to 1) qualitative assessment of species richness patterns in the survey data, 2) scale dependency analysis of richness and a suite of landscape variables, and 3) correlation analysis between richness and landscape variables. We define richness as the total number of unique species collected at each survey site.

Plotting Species Richness

Our first step in assessing spatial patterns of species richness was to plot, for each taxonomic group and subgroup, site richness over land cover. Graduated circles, proportional to species numbers, help visualize richness at each site (see Figure 1 for example). These maps provided a quick analysis of spatial patterns of species richness and whether these patterns appeared coincident to land cover.

Map of Butterfly Species Richness

Scale Dependency Analysis

The second step of our analysis focused on defining spatial scale dependencies of both the response variable, richness, and the independent variables; slope, aspect, insolation, elevation, and land cover. Species richness was analyzed by plotting the variogram of richness at increasing distance using the UTM coordinates of site centroids. MGAP geostatistical software was used to calculate the variogram. (See Figure 2a). The variogram is a powerful tool for looking at the change in variability of a measured attribute with increasing distance, or its degree of spatial autocorrelation. When a collection of samples taken close together are more similar than samples taken at greater distances, the closer samples are said to be positively autocorrelated. Spatial autocorrelation is believed, therefore, to introduce statistical bias (pseudo replication) if sample sites are not far enough apart. Samples are not spatially autocorrelated when the variogram shows an asymptotic response to increasing distance (lag).

Semivariogram of Butterfly Richness

Landscape variables were checked for spatial autocorrelation using a different statistic, the Moran Index (Goodchild 1986), which is provided as a function within GRID. This index ranges between -1 (strongly negatively autocorrelated) and 1 (strongly positively autocorrelated). Since the Moran Index is calculated only from adjacent grids (lag equals cell resolution), we calculated, for each variable, a series of grids at incrementally larger cell resolutions (35m, 50m, 100m, 200m, 300m,Ö1000m). Each series of grids was generated by resampling the original raster (28.5 meter resolution) at the specified resolution. Results are presented in Figure 3.

Moran Index for Five Environmental Variables

Correlation Analysis

Quantitative analysis used buffers and overlays to generate landscape variables at incremental distances around each site. Most GIS processing was performed in GRID. A zonal grid defining distance class was generated for each site by rasterizing a multi-buffer polygon coverage containing concentric circles around each site centroid at 100 meter intervals out to one kilometer as well as smaller buffers at 35 and 50 meters. Zonal grid cells were 28.5 meters on a side to maintain spatial parity among inputs.

The zonal distance grids were then ìlayered overî terrain and land cover rasters using zonal functions to derive new grids relating neighborhood size with landscape attributes. The value attribute table (VAT) from these zonal overlays were first converted to integer format, if required, then unloaded into a spreadsheet for additional data manipulation and statistical analysis.

Initial analysis of Ithomiine butterfly data suggested that species richness was related more to the habitat, specifically, the proportion of primary forest, surrounding a sample site (neighborhood) than the habitat type in which the sample was taken (Levy 1995). This observation prompted us to generate the area of primary forest within each distance class around each site. Two other derivatives of land cover were calculated to explore the influence of landscape heterogeneity (landscape diversity index) and fragmentation (edge/area). Landscape diversity was calculated using Shannon-Wienerís diversity index (Ricklefs 1979:686). Cumulative proportions of each land cover class at each distance interval were used to calculate landscape diversity. Edge per unit area was calculated by first converting the overlay output of zonal distance and land cover into a polygon coverage. An automated routine was scripted in AML to extract vectors out to each distance class and dissolve by land cover type. A frequency count was run on the resulting vector coverage providing both the number and total length of edge between land cover classes. Total edge per distance class per site was divided by the total area calculated in the zonal distance grid.

Of all the terrain variables calculated, we only tested insolation range as a predictor of species richness. Insolation captures variability in slope and aspect suggesting it may be a better predictor of habitat heterogeneity than either of the individual terrain variables.

To explore the spatial scale of landscape and physiographic variables that best predicted species richness, we regressed percent primary forest, landscape diversity, edge/area, and insolation range against species richness within each taxonomic group at each distance class. The taxonomic groups tested included all butterflies, Satyrine butterflies, Ithomiine butterflies, Nymphalid butterflies, Eluethrodactylid frogs, Hylid frogs, understory birds and forest interior shrub-layer birds. Several transformations of the input data were employed to reduce estimate bias and increase normality. Species richness was log transformed while percent primary forest was transformed by taking the square root of the arcsine. JmpIn statistical software was used to calculate all simple regressions. R2 values were plotted against neighborhood sizes to determine what size best explained observed richness patterns.

Discussion

The maps of species richness symbolized over land cover confirmed field observations that overall richness of butterflies and understory birds exhibited non-random distributions relative to land cover. Maps of taxonomic subsets which we wanted to assess as indicators of richness varied in the appearance of distribution patterns suggesting weak or insignificant correlations between indicator groups of butterflies and frogs and the larger taxonomic groups to which they belong. Forest interior shrub layer birds did show obvious distribution patterns similar to the larger sample of all non-canopy birds.

The variogram of overall butterfly richness presented in Figure 2a appears to sill between 400 and 500 meters. Semivariance starts to increase again at 1200 meters suggesting species richness may exhibit patterns at multiple scales. One might even speculate that the first sill represents the scale at which alpha diversity in butterflies in this region varies. With additional data from sites at greater distances we might expect to see a second sill in the variogram at lag distances beyond 3000 meters representing the spatial scale of beta diversity within this region. Such speculation is compelling but the results are tenuous given the unreliably small sample sizes used to calculate semi-variance (Figure 2b).

A detailed description and discussion of correlation results are presented elsewhere (Pearman et. al., submitted). However, we can report that a common trend did appear among all four independent variables tested for several of the taxonomic groups. Principally, species richness was found to be most significantly and strongly predicted at medium distances beyond the immediate zone of each sample site. The biological interpretation of these results is intuitive given the vagility of the organisms being studied.

In conclusion, our data show that the ìneighborhood countsî, e.i., the number of species found at any one site is more representative of the environment around the that site than by the environmental conditions where the sample is drawn. With specific data on neighborhood sizes, or spatial scale of taxonomic richness, for several different taxonomic groups, we arrive at simple guidelines for further analysis of our data set and for designing future studies of species distribution in this region.

References

Allen, T.F.H. and T.B. Starr. 1982. Hierarchy, perspectives for ecological complexity. University of Chicago Press, Chicago.

Davis, F.W., D.A. Quattrochi, M.K. Ridd, N.S. Lam, S.J. Walsh, J.C. Michaelsen, J. Franklin, D.A. Stow, CJ. Johannsen, and C.A. Johnston. 1991. Environmental analysis using integrated GIS and remotely sensed data: some research needs and priorities. Photogrammetric Enginnering and Remote Sensing, 57:689-697.

Ecological Society of America. 1991. The sustainable biosphere initiative: an ecological research agenda. Ecology 72:371-412.

Gaston, K.J. 1996. Spatial covariance in the species of higher taxa. Pages 221-242 in M.E. Hochberg, J. Colbert and R. Barbault (eds.). Aspects of the Genesis and Maintenance of Biological Diversity. Oxfor University Press, Oxford.

Goodchild, M.F. 1986. Spatial autocorrelation. Catmog 47, Geo Books, Norwich.

Hetrick, W.A., Rich, P.M., Barnes, F.J., and S.B. Weiss. 1993. GIS-based solar radiation flux models. American Society for Photogrammetry and Remote Sensing Technical Papers 3(Photogrammetry and Modelling): 132-143.

Instituto Geogr·fico Militar. 1988. Puerto MisuhuallÌ. 1:50,000 Map. Quito, Ecuador.

Isaaks, E. H. and R. M. Srivastava. 1989. An introduction to applied geostatistics. Oxford University Press, Oxford.

Levy, K. 1995. The utility of Ithomiinae (Lepidoptera: Nymphalidae) as indicators of forest disturbance: a case study of the applicability of biological research to conservation. Goldman Honors Thesis. Environmental Science, Technology, and Policy. Stanford University.

Noss, R.F. 1990. Indicators for monitoring biodiversity: a hierarchical approach. Conservation Biology 4:355-364.

Pearman, P.B., A.M. Velasco, and A. López. 1995. Tropical amphibian monitoring: a comparison of methods for detecting inter-site variation in species composition. Herpetologica 51:325-337.

Pearman, P.B., T.D. Sisk, M.Guerrero-Gutierrez, H.R. Sparrow, and C.L. Boggs. In Press. The value of correlated patterns of tropical animal diversity for guiding reserve selection and management.

Pearman, P.B., C.L. Boggs, M.G. Guerrero, B.C. Sandler, K. Levy. Submitted.

Pearson, D.L., and F. Cassola. 1992. World-wide species richness patterns of tiger beetles (Coleoptera: Cicindelidae): indicator taxon for biodiversity and conservation studies. Conservation Biology 6:376-391.

Prendergast, J.R., R.M. Quinn, L.H. Lawton, B.C. Eversham, and D.W. Gibbons.

Ricklefs, R.E. 1979. Ecology. Chiron Press Inc, New York, p. 686.

Webster, R. and M.A. Oliver. 1990. Statistical Methods in Soil and Land Resource Survey. Oxford University Press, Oxford.

Wilson, E.O. 1988. The current state of biological diversity. Pages 3-18 in E.O. Wilson and F.M. Peter, editors. Biodiversity. National Academy Press, Washington, D.C.


Bennett Sandler
GIS Lab, Center for Conservation Biology
Stanford University
Stanford, CA 94309
PH: (415) 725-5585
FAX: (415) 723-5920
email: bennett@bing.stanford.edu

Peter Pearman, PhD.
The Evergreen State College
Olympia, WA 98505
PH: (360) 866-6000
FAX: (360) 866-6794
email: pearmanp@elwha.evergreen.edu

Mauricio Guerrero G.
Centro para la Biologia de la Conservacion
Programa de Investigation Tropical - Ecuador
Av. Rio Coca 1734
Casilla Postal: 17-03-751 Quito - Ecuador
Telfax: (593-2) 441-592 / 250-976
email: mguerrer@jsacha.ecx.ec

Karen Levy
Environmental Defense Fund
Rockridge Market Hall
5655 College Ave.
Oakland, CA 94618
PH: (510) 658-8008
FAX: (510) 658-0630
email: karenl@edf.org