 Figure 1: The Maya Biosphere Reserve and its buffer zones (ZAM and ZUM). |
Modeling Deforestation Risks for the Maya Biosphere Reserve, Guatemala |
Wolfgang Grunberg, D. Phillip Guertin, and William W. Shaw
School of Renewable Natural Resources, The University of Arizona, Tucson,
Arizona, USA
The tropical forest of Guatemala's 21,130 square kilometer Maya Biosphere Reserve and buffer zone is being impacted by extended deforestation due to an increase of the local population and establishment of over 200 new settlements over the last 30 years. Existing GIS and remote sensing data were used to determine how much of the observed deforestation could be explained by three factors: roads, human settlements, and soil quality. Each factor was modeled and tested separately using spatial and statistical analysis methods, and were then combined to create a final deforestation risk model. The deterministic model enables policy makers, as well as managers, to create scenarios that assess the impact of their actions on the forest on a regional scale.
Deforestation of tropical forests is a widespread management and conservation issue throughout the tropics. In the case of the tropical lowland forest in the northern Petén, Guatemala, national and international conservation and development agencies have collected biophysical and socio-economic data that documents the area’s deforestation trends and factors over the past years. The municipality of Petén, located in northern Guatemala, contains one of the largest continuous tropical forests of Central America. In 1990, 40% of the forested area (21,130 square kilometer; CONAP et al., 1996) was set aside as the Biósfera Maya Nature Reserve (MBR) with a surrounding buffer zone (Zona de Amortiguamiento - ZAM) and multiple use zone (Zona de uso Múltiple - ZUM; Figure 1). However, the dramatic increase of the Petén's population through immigration of various Mayan peoples and Ladinos into the region and establishment of more than 200 new settlements over the last 20 years led to extended and rapidly increasing deforestation of the Petén including the reserve (Grünberg and Ramos, 1998). The settlements' subsistence economy - based on ranching and traditional farming - led not only to well-documented loss of bio-diversity but also to economic loss due to soil degradation (CONAP et al., 1996).
Little is known regarding specific environmental, as well as socio-economic factors, that contributed to deforestation of the MBR, ZAM, and ZUM. Quantifying the importance of a factor to assess the risk of deforestation would enable policy makers, as well as managers, to create scenarios and model the impact of their actions on the forest on a regional scale. This paper will discuss the influence of three factors: roads, human settlements, and soil quality on deforestation and how those factors can be used to model deforestation.
The Petén is comprised of a hilly, limestone karst landscape with an elevation from ca. 100 to 300 m above mean sea level (Islebe et al. 1996). Its mean annual temperature is 25° C and the region's precipitation averages 1600 mm a year (Islebe et al. 1996). The region's vegetation zones consist of the predominating high canopy tropical lowland forest, occasionally inundated lowland areas or bajos of chaparral-like dense shrubs and small trees, wetlands along rivers and around lakes, and flat savanna-like grasslands.
The area remained sparsely populated for most of the last 500 years by the native lowland Mayan peoples and Ladinos, who traditionally practiced agroforestry (Atran 1993). Modern settlements base their subsistence economy on ranching and maize swidden farming (Santiso 1993; Schwartz 1990).
Originally, the Itza Maya population moved in from the Yucatan in the north before the Spanish conquest into the area that is now part of the MBR's buffer zone (Atran 1993). The Itza, who rely heavily on forest products, are not considered to be related to the ancient Mayan civilizations that built the City States which, now in ruins, are famous archeological and tourist attractions for the region. Today, the small Itza communities rely on growing corn, wage labor, hunting, and collecting xate (Chamaedorea elegans) - a palm whose ornamental leaves are exported to the U.S and Europe.
The Peteneros, a small local non-immigrant population of Hispanic descent, have lived in the northern Petén and the MBR since the Spanish colonization period. Traditionally, the Peteneros have been close to the Itza and adopted much of their knowledge and agroforestry techniques such as home-gardens, rotational swidden farming with multi-cropping, and forest gardens (Atran 1993). Although agroforestry, hunting, and xate collecting are also prevailing income and food sources, the Petenero are currently shifting their economic mode to wage labor, commerce, and ranching.
The immigrant Mayan peoples such as the Q'eqchi' and Kaqchikel entered the Petén in the last 30 years from the south due to socio-economic pressures such as a population boom in their original communities, limited access to land, and persecution (Georg Grünberg, personal communication). Similar to the Itza, their main staple is corn. However, their agricultural techniques are mostly adapted to the often volcanic soils and more moderate climate of the highlands of central Guatemala and not to the fragile soils of the Petén. Although the older immigrant communities are notably adapting their agriculture techniques to the new environmental conditions, soil degradation, pests, and low yields are common.
The large majority of immigrants into the Petén are of mixed Hispanic and Mayan descent - called Sureńos or Ladinos - and have been moving into the MBR and its buffer zone since ca. 20 years for mostly economic reasons (Grünberg and Ramos 1998). Similar to the highland Mayans, recent Ladino immigrants rely on subsistence farming of corn and beans using swidden-fallow agriculture and suffer from the same mal-adapted agricultural techniques. However, Ladinos tend to introduce small scale ranching as soon as they can afford it.
The study’s GIS database, compiled from various sources and formats, was standardized to the UTM projection, Zone 16, units in meters, and NAD Central America datum. The different vector and raster themes had a root mean square error (RMS) of approximately 400 meters to each other (Thapa and Bossler 1992).
The foundation of this study was a series of forest cover change detection images for nearly the entire MBR, ZUM, and ZAM developed by the Maine Image Analysis Laboratory, University of Maine. The change-detection images were derived from mosaiced Landsat Thematic Mapper satellite images based on a normalized difference vegetation index (NDVI = [near infrared – red]/[near infrared + red]; Sader et al. 1997, 1998). The images displayed forest cover changes at a 30 by 30 meter resolution for the first and second quarter of 1986, 1990, 1993, 1995, 1997, and 1999. They represented a cumulative deforestation process and did not account for forest regrowth. The 1990 to 1997 forest cover change detection images had an overall classification accuracy of 86.46% (Sader, personal communication). The 1999 image, for the 1997 to 1999 period, did not cover the entire study area due to cloud-cover and the forest cover change detection was hindered by widespread man-made wildfires caused by slash-and-burn forest clearing techniques. The 1986 to 1997 images, referred to as the analysis period, were used to create the model, while the 1999 image was used for testing the 1999-risk model.
Spatial and socio-economic data for 194 settlements up to March, 2000, were provided by CARE Guatemala and the Centro de Monitoreo y Evaluacion del Consejo Nacional de Areas Protegidas (CEMEC-CONAP; Grünberg and Ramos 1998, personal communication). Census data for the settlements was not used in the deforestation analysis because they were inadequate or not present. An electronic 1:200,000 Food and Agricultural Organization of the United Nations (FAO) soil series map provided by CONAP was reclassified to drainage and soil depth categories according to FAO agricultural suitability guidelines and local agronomist Vinicio Montero (Table 1; personal communication). Temporal and spatial roads and bus data were compiled from information provided by Perfecto Carillo, CONAP-CEMEC, Marco Antonio Palacios, and the Wildlife Conservation Society Gainesville (WCS; personal communication). CONAP-CEMEC and WCS also provided spatial data on administrative boundaries, rivers, and lakes.
Twenty concentric 1-km-diameter rings at increasing distances from the center buffered each individual settlement point. Next, the percent deforestation of each 1-km ring and analysis year was calculated. The percent deforestation data of the settlements was averaged according to socio-economic categories such as major economic mode and major ethnicity to identify variations in deforestation trends caused by socio-economic settlement differences. The generalization of the socio-economic categories, based on observations of anthropologists Georg Grünberg and Norman Schwartz, were necessary to increase the sample sizes (personal communication).
The reclassified soil quality map and the deforestation images were used to calculate the percent deforestation within the area of each soil quality category during the analysis period to determine the relationship between deforestation trends and soil quality.
Only paved and dirt roads that were passable all year long with regular pickup tracks or public buses were included in the analysis. Intermittent or temporary dirt roads were not included because of their unknown status or extent. Also, a well maintained stretch of the "ruta petrolera" from the town El Naranjo by the San Pedro river to the North was not included until 1991 because access was controlled through a ferry by oil companies (BASIC). It was assumed that the entire study area was easily penetrated on foot, with mules, or four-wheel drive pickup trucks during most of the year except during the rainy season in summer. Perennial roads, however, are significant to settlements because they allow for cheap transportation, such as public buses, and easy access for mobile merchants and their trucks.
For the descriptive model, a cell by cell logistic regression was calculated for each analysis year. The dependent variable was the binary forested/deforested grid and the independent variables were grids representing well/poorly draining soils, natural log transformed distance to the closest road, and natural log transformed distance to the closest settlement. Due to the study area's large size, 5% stratified random samples (> 1,100,000 cells) were used for the statistical analysis - where 5% of the forested cells and 5% of the deforested cells were sampled independently to be combined later (Appendix). The stratification of the samples was necessary in order to ensure the approximately 1:25 ratio of deforested to forested cells. In addition, the y-intercepts or constants of the logistic regressions were corrected for unequal sample sizes (Warren, 1990; Appendix).
The independent variable grids were then multiplied by their respective regression coefficient and the corrected y-intercept was added as a constant. Finally, the sums of weighted grids were logistically transformed to create probability surfaces with values between 0 (lowest deforestation probability) and 1 (highest deforestation probability) for each analysis year (Figure 2; Appendix). All probability models were compared with their respective forest change detection images.
First, the regression coefficients for the 1999 deforestation probability prediction were estimated by fitting regression curves to the coefficients of the analysis period and by projecting the curves to 1999. To account for the variation of the regression coefficients, 75 0.25% samples were taken resulting in 15 regression coefficients for each regression variable and analysis year. The predicted 1999 probability model was then calculated by weighing the 1999 soil and distance grids by the projected coefficients (Figure 2). The total deforestation area for 1999 was estimated by fitting a regression curve on the observed total deforestation for each analysis year and by projecting the curve to 1999. Next, a regression curve was fitted to the observed distributions of deforestation along the probability zones of all analysis years. The generalized deforestation distribution curve was used to distribute the deforestation estimate for 1999 along the probability zones of the predicted 1999 probability model to create the final 1999 deforestation risk model. The projected deforestation probability and risk models were compared with the 1999 change detection image for testing purposes.
All spatial and statistical analyses were performed with the SPSS statistics package and the ARC/INFO and ArcView GIS software on Windows NT and Solaris workstations.
The settlements of all economic modes but wage labor had, on average, a noticeable deforestation impact on their surroundings at least to a distance of 4 kilometers (Figure 3). Wage labor settlements had little impact on the immediate forest and were, therefore, removed from the deforestation probability models. The deforestation distance decay curves of settlements grouped according to their major ethnicity showed differences in behavior but were not believed to be significant enough to be included in the models (Figure 4). All non-settlement sites were supported by wage labor and were consequently excluded from the models. In addition, CPR guerrilla settlements in the Sierra del Lacandón National Park were not included in the model because of their minimal impact on the forest in order to maintain cover.
The simplified agricultural soil usability map showed that well draining soil types were more likely to be deforested than poorly draining soils regardless of their depth (Figure 5). In addition, over time, the deforestation rate increased on well draining soils while it remained close to 0 on poorly draining soils (Figure 5). Because of the observed deforestation trends and the soil map's poor resolution, the soil variable for the probability model was simplified to the binary categories of well draining (1) and poorly draining soils (0) (Table 1).
Bus routes with regular public transportation were poor deforestation probability indicators compared to perennial roads. The bus routes were therefore disregarded in favor of perennial roads for the probability models.
As expected, the deforestation probability models did not predict the amount of observed deforestation (Figure 6) but were good indicators of deforestation distribution (Figure 7). Consistently for the analysis years, the higher the deforestation probability score, the higher was the probability zone's observed deforestation rate (Figures 7-17). This indicated that the chosen soil and distance variables captured the likelihood of deforestation.
Most of the regression curves for the coefficient time series did not fit well and made projections for 1999 highly subjective. Except for the soil drainage coefficient, the other coefficients’ variations may imply no actual change over time. Nevertheless, the projected 1999 coefficients were derived from the best fitting regression curves (Figures 18-21).
The regression curve for the total deforestation time series fitted well (Figure 22). Although the quadratic curve showed a large confidence interval, it was used to project the total area of deforestation for 1999 (95% CI = 2380 square kilometer ± 420).
The percent deforestation distributions among the deforestation probability zones for 1985 to 1997 were very similar (Figure 23). A quadratic regression curve was fitted to observed distributions in order to model the deforestation distribution for the 1999 deforestation risk model (Figure 23).
Although human-made wildfires in the 1997 to 1999 period - especially during the dry season of March, April and May of 1998 - complicated a comparison, the distribution of observed and predicted percent deforestation in 1999's predicted deforestation probability zones were similar to each other (Figure 24). The large difference between observed and predicted percent deforestation for the 0.95-1 probability zone could be explained by the ill fit of the quadratic curve. The distribution of all the analysis years showed a similar trend where the percent deforestation increaseed exponentially to the 0.9 probability mark and then flattened out (Figure 23). This could be in part an artifact of the spatial data’s 400-meter RMS error or an artifact of the logistic regression model and the y-intercept correction.
The final 1999 deforestation risk model was based on the projected 1999 deforestation probability model (Figures 24, 26, and 27), quadratic deforestation distribution curve (Figure 23), and projected 1999 total deforestation (Figure 22). It was relatively close to the observed deforestation with an overall error of 15.08% (Figure 25, Table 2). In order to compare predicted and observed deforestation, the 1999 total deforestation projection was adjusted for decreased observation area in 1999 (Figure 26). The differences in observed and predicted deforestation may have reflected the differences between the observed distribution for 1986 to1997 and the quadratic deforestation distribution curve used for the final risk model (Figures 23 and 25).
 Figure 26: MBR’s settlements, roads, cumulative deforestation, and areas impacted by man-made wildfires (mainly during the dry season of March, April, and May, 1998) in 1999.
|
Several studies (Chomitz and Gray no date; Mas et al. 1997; Mertens and Lambin 1997;) have shown a clear relationship between the presence of roads and human settlements and deforestation. As expected, the results of this study supports these findings. This study shows that soil quality and natural log transformed distances for roads and settlements are good indicators for deforestation trends in an agricultural frontier such as the Petén.
The deforestation risk model does not predict deforestation locations per se but can be used to estimate the possible impact of new roads and settlements. For example, it can be used as an environmental impact assessment tool for a proposed road through the MBR to Mexico or other scenarios (WCS- Petén, personal communication).
The deforestation model does not account for spatial or temporal autocorrelation. The analysis methods used in this study may have been successful on a regional scale, but better methods that combine spatial and temporal analysis are needed for modeling deforestation on a local level. For example, such an analysis method would need to account for roads which have segments of different ages that "grow" over time and it would need to quantify the overlapping effect of those road segments. Panel data or pooled time data analysis may be more adequate methods.
Although in this study it may have been proven useful, using regression curves for estimating values beyond the samples’ range is statistically questionable and remains a subjective method. The projected estimates will of course improve as more years are added to the analysis period. As a test projection of the deforestation probability for the year 2001 showed, any projection further than 2 years will likely fail.
The deforestation risk model’s greatest strength is its simplicity as a general model with relatively good results. The existence of adequate census data might have improved the deforestation risk model significantly but their lack did not hinder it - which is advantageous for modeling in frontier regions which often lack census data. Once the coefficients are calculated, a deforestation probability or risk model can be relatively easily calculated using map algebra on common spatial features such as roads, settlement points, and simple soil maps. As a general model, it may prove to be useful not only for the northern Petén but also for other agricultural frontiers.
This study led to the following, often obvious, suggestions regarding reducing deforestation risks:
The study’s analysis has room for improvement. For example, the 1999 observation may be incorporated for a 2001 deforestation risk projection. The model needs to be applied in scenarios of managerial significance, such as that of the proposed road to Mexico. Spatial and temporal autocorrelation need to be accounted for. Water availability for ranching and agriculture could be incorporated. River traffic and oil-pipelines need to be considered as access routes and settlements may be differentiated according to their socio-economic qualities. A simple stochastic component could be incorporated into the model for better visualization of deforestation risk. Also, the model may have uses for estimating areas of high wildfire risks. Finally, slope and aspect data of adequate resolution in combination with better soil maps may turn this regional model into a more localized version.
The authors would like to thank the following organizations and individuals for their indispensable help: Advanced Resource Technology Group, CARE Guatemala, CONAP-CEMEC, CI-ProPeten, WCS-Gainesville, Perfecto Carillo, Teresita Chinchilla, Gary Christopherson, Reno Fiedler, Georg Grünberg, Vinicio Montero, Randy Gimblett, Gustavo Rodriguez Ortiz, Marco Antonio Palacios, Victor Hugo Ramos, Steven Sader, Claudio Saito, Norman Schwartz, Carlos Soza, and Craig Wissler.
Figure 3: Deforestation distance decay curves of settlements according
to their main economic mode.
Figure 4: Deforestation distance decay curves of settlements according
to their ethnic majority.
Figure 5: Deforestation trends on various soil classes.
Figure 6: Observed deforestation and deforestation probability for 1997
(0.5 cutoff point).
Figure 7: Observed deforestation distribution within predicted probability
zones for the year 1997.
Figure 8: MBR’s settlements, roads, and cumulative deforestation
in 1986.
Figure 9: Deforestation probability model for 1986.
Figure 10: MBR’s settlements, roads, and cumulative deforestation
in 1990.
Figure 11: Deforestation probability model for 1990.
Figure 12: MBR’s settlements, roads, and cumulative deforestation
in 1993.
Figure 13: Deforestation probability model for 1993.
Figure 14: MBR’s settlements, roads, and cumulative deforestation
in 1995.
Figure 15: Deforestation probability model for 1995.
Figure 16: MBR’s settlements, roads, and cumulative deforestation
in 1997.
Figure 17: Deforestation probability model for 1997.
Figure 18: Corrected y-intercept coefficient 1986-97 (75 0.25% samples,
quadratic curve fit, R^2 = 0.6292, 95% confidence interval of the mean).
Figure 19: Natural log transformed road distance coefficient 1986-97
(75 0.25% samples, quadratic curve fit R^2 = 0.5660, 95% confidence interval
of the mean).
Figure 20: Natural log transformed settlement distance coefficient 1986-97
(75 0.25% samples, cubic curve fit, R^2 = 0.7414, 95% confidence interval of
the mean).
Figure 21: Soil drainage coefficient 1986-97 (75 0.25% samples, cubic
curve fit, R^2 = 0.9420, 95% confidence interval of the mean).
Figure 22: Total Deforestation 1986-97 and projection estimate (quadratic
curve fit 95%, R^2 = 0.9936, 95% confidence interval of the mean).
Figure 23: Total deforestation distribution 1986-97 (quadratic curve
fit, R^2 = 0.9773, 95% confidence interval of the mean).
Figure 24: Observed vs. predicted deforestation distribution of the 1999
prediction model. Most of the fire occurred during the dry season of March,
April, and May, 1998.
Figure 25: Observed vs. predicted deforestation area of the 1999 deforestation
risk model. The predicted deforestation was adjusted for the decrease of the
1999 observation area due to cloud cover in the Sierra del Lacandón National
Park area. Most of the fire occurred during the dry season of March, April,
and May, 1998.
Table 2: Differences between predicted and observed
deforestation for 1999. The predicted deforestation was adjusted for the decrease
of the 1999 observation area due to cloud in the Sierra del Lacandón National
Park area.
Deforestation Probability Score |
Observed Deforestation (km^2) |
Predicted Deforestation (km^2) |
% Difference |
|---|---|---|---|
0 - 0.05 |
0.5 |
2.5 |
361.38 |
0.05 - 0.10 |
6.1 |
6.9 |
12.86 |
0.10 - 0.15 |
1.3 |
12.2 |
872.66 |
0.15 - 0.20 |
0.8 |
18.6 |
2148.09 |
0.20 - 0.25 |
6.1 |
26.5 |
333.94 |
0.25 - 0.30 |
13.7 |
36.6 |
168.03 |
0.30 - 0.35 |
21.5 |
47.8 |
121.88 |
0.35 - 0.40 |
25.3 |
60.1 |
137.05 |
0.40 - 0.45 |
42.0 |
73.7 |
75.50 |
0.45 - 0.50 |
76.1 |
88.5 |
16.32 |
0.50 - 0.55 |
97.0 |
104.6 |
7.90 |
0.55 - 0.60 |
124.0 |
121.9 |
-1.66 |
0.60 - 0.65 |
157.9 |
140.6 |
-10.95 |
0.65 - 0.70 |
170.6 |
160.8 |
-5.78 |
0.70 - 0.75 |
195.4 |
181.5 |
-7.11 |
0.75 - 0.80 |
217.0 |
204.0 |
-5.99 |
0.80 - 0.85 |
227.1 |
227.4 |
0.15 |
0.85 - 0.90 |
236.8 |
252.1 |
6.44 |
0.90 - 0.95 |
243.8 |
277.5 |
13.78 |
0.95 – 1 |
175.8 |
302.7 |
72.18 |
Total Deforestation |
2038.9 |
2346.4 |
15.08 |
1. Statement used to create a stratified 5% random sample grid of the study area’s deforestation grid:
In ARC/INFO's GRID module:
out_grid = con(in_grid eq 1, con(rand (.) < 0.05, 1, in_file eq 0, con(rand (.) < 0.05, 0)
Where 0.05 is the proportion of the total number of cells in the grid that are desired as deforested (1) or forested (0) cells.
2. Formula for calculating the corrected y-intercept or the constant regression coefficient. Correcting the y-intercept was necessary to account for the unequal sample sizes between deforested (0) and forested (1) cells.
a’ = a + ln (n2/n1)
Where a is the y-intercept in the regression, a’ is the corrected y-intercept, ln is a natural logarithm, n1 is the number of cases in the smaller sample (deforested cells), and n2 is the number of cases in the larger sample (forested cells). This equation follows Warren , R. E. 1990.
3. Formula for the logistic transformation of the probability surfaces:
In ARC/INFO’s GRID module:
docell
temp1:= exp (-1 * in_grid)
out_grid = 1/(1 + temp1)
end
Atran, Scott. 1993. Itza Maya tropical agro-forestry. Current Anthropology. 34(5):633-700.
Chomitz, Kenneth M. and David A. Gray. No date. Roads, land, markets and deforestation - a spatial model of land use in Belize. Environmental, Infrastructure, and Agriculture Division Policy Research Department, PERDEI, The World Bank. 50pp.
CONAP, USAID and Peregrine Foundation. 1996. El estado de la Reserva de la Biosfera Maya en 1996. CONAP, Petén, Guatemala. 27pp.
Grünberg, Georg and Victor Hugo Ramos. 1998. Base de Datos sobre poblacion, tierras y medio ambiente en la Reserva de la Biosfera Maya. CARE Guatemala and CONAP. Petén, Guatemala. 91pp.
Islebe, Gerald A., Henry Hooghiemstra, Mark Brenner, Jason H. Curties and David A. Hodell. 1996. A Holocene vegetation history from lowland Guatemala. The Holocene. 6:265-271.
Mas, Jean-Francoise, Azucena Perez Vega, Gerardo Palacio Aponte, and David Zarate Lomeli. 1997. Assessing deforestation in the coastal zone of Campetche State, Mexico. Proceeding of the IV International Conference on Remote Sensing for Marine Coastal Environments. Orlando, Florida. Vol. 1 pp 593-602.
Mertens, Benoit, and Eric F. Lambin. 1997. Spatial modelling of deforestation in southern Cameroon spatial disaggregation of diverse deforestation processes. Applied Geography. Vol. 17. No. 2. pp 143-162.
Sader, Steven A., Conrad Reining, Thomas Sever, and Carlos Soza. 1997. Human migration and agricultural expansion. Journal of Forestry. Vol. 95. No. 12. pp. 27-32.
Sader, Steven A., Michael Coan, and Daniel Hayes. 1998. Time-series tropical forest change detection for the Maya Biosphere Reserve: Updated Estimates for 1995 to 1997. Maine Image Analysis Laboratory, Department of Forest Management, University of Maine. Unpublished report. CI-ProPeten. Petén, Guatemala.
Santiso, Claudio. 1993. The Maya Biosphere Reserve: an alternative for the sustainable use of resources. Nature and Resources. 28(1-4):6-11.
Schwartz, Norman B. 1990. Forest society - a social history of Petén, Guatemala. University of Pennsylvania Press, Philadelphia. 367pp.
Thapa, Khagendra and John Bossler. 1992. Accuracy of spatial data used in geographic information systems. Photogrammetric Engineering & Remote Sensing. Vol. 58. No. 6. pp. 835-841.
Warren, R. E. 1990. Predictive modeling of archeological site location: a case study in the Midwest. P210-215 in Interpreting space: GIS and archeology. London, Taylor and Francis.
Wolfgang Grunberg is a Renewable Natural Resource Sciences Graduate Student, School of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721, USA. grunberg@u.arizona.edu
D. Phillip Guertin is an Associate Professor in Watershed Management and Landscape Studies, School of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721, USA.
William W. Shaw is a Professor in Wildlife and Fisheries Sciences, School of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721, USA.
