Western Kentucky University
TopSCHOLAR®
Masters Theses & Specialist Projects
Graduate School
12-1-2013
Eastern Deciduous Forest Phenology and
Vegetative Vigor Trends From 2000 to 2013,
Mammoth Cave National Park, KY
Sean Taylor Hutchison
Western Kentucky University, [email protected]
Follow this and additional works at: http://digitalcommons.wku.edu/theses
Part of the Geographic Information Sciences Commons, Geology Commons, and the Physical
and Environmental Geography Commons
Recommended Citation
Hutchison, Sean Taylor, "Eastern Deciduous Forest Phenology and Vegetative Vigor Trends From 2000 to 2013, Mammoth Cave
National Park, KY" (2013). Masters Theses & Specialist Projects. Paper 1312.
http://digitalcommons.wku.edu/theses/1312
This Thesis is brought to you for free and open access by TopSCHOLAR®. It has been accepted for inclusion in Masters Theses & Specialist Projects by
an authorized administrator of TopSCHOLAR®. For more information, please contact [email protected].
EASTERN DECIDUOUS FOREST PHENOLOGY AND VEGETATIVE VIGOR
TRENDS FROM 2000 TO 2013, MAMMOTH CAVE NATIONAL PARK, KY
A Thesis
Presented to
The Faculty of the Department of Geography and Geology
Western Kentucky University
Bowling Green, Kentucky
In Partial Fulfillment
of the Requirements for the Degree
Master of Science
By
Sean Taylor Hutchison
December 2013
Dedication
For my mother and father.
ACKNOWLEDGMENTS
I would like to thank my advisor Dr. John All for providing the academic support
to help me complete this project. My thesis committee and the faculty of the Department
of Geography and Geology at Western Kentucky University are acknowledged for their
guidance and helpfulness. I appreciate the cooperation and helpfulness of Steven Thomas
and Teresa Leibfreid of the National Park Service, particularly for their financial support.
I thank Dr. Ken Kuehn for his support and assistance beginning this journey. I thank
Sarah Rehkopf Schoefernacker for taking me under her wing and teaching me about the
fascinating world of Geography. Finally I would like to thank my wife, Sara Hutchison,
for her unending patience and constant encouragement.
iv
CONTENTS
LIST OF FIGURES .......................................................................................................... vii
LIST OF TABLES ............................................................................................................. ix
Abstract ............................................................................................................................... x
Chapter 1: Introduction ....................................................................................................... 1
Chapter 2: Background ....................................................................................................... 5
2.1 Climate Change Impacts on Forest Ecology............................................................. 5
2.2 Study Area ................................................................................................................ 7
2.2.1 Climate ............................................................................................................... 8
2.2.2 Land Cover and Geology ................................................................................. 10
2.2.3 Vegetation Classifications ............................................................................... 14
Chapter 3: Geospatial Tools and Climate Indices............................................................. 17
3.1 Remote Sensing ...................................................................................................... 17
3.1.1 Normalized Difference Vegetation Index ........................................................ 19
3.1.2 Data Acquisition and Processing ..................................................................... 21
3.2 Teleconnections ...................................................................................................... 26
3.2.1 Multivariate ENSO Index ................................................................................ 26
3.2.2 Pacific Decadal Oscillation .............................................................................. 27
3.2.3 Pacific/North American ................................................................................... 29
3.2.4 North Atlantic Oscillation ................................................................................ 31
v
Chapter 4: Methods ........................................................................................................... 33
Chapter 5: Results and Discussion.................................................................................... 38
5.1 Aggregated NDVI ................................................................................................... 38
5.2 Phenology Shifts ..................................................................................................... 45
5.3 NDVI and Teleconnections..................................................................................... 50
5.4 Data Interpretation .................................................................................................. 55
Chapter 6: Conclusions ..................................................................................................... 60
APPENDIX A: RASTER PROCESSING SCRIPTS ....................................................... 63
Literature Cited ................................................................................................................. 69
vi
LIST OF FIGURES
Figure 1. Mammoth Cave National Park Land Cover and Regional Context. ................. 11
Figure 2. Generalized Land Cover Classes Surrounding Mammoth Cave National Park.12
Figure 3. Generalized geologic map of Mammoth Cave National Park. .......................... 13
Figure 4. Generalized Vegetation Classes at MACA. ...................................................... 16
Figure 5. Normalized Difference Vegetation Index (NDVI) Diagram. ............................ 20
Figure 6. Extent of Single MODIS tile. ............................................................................ 24
Figure 7. Multivariate ENSO Index (MEI) 2000 to 2013. ................................................ 28
Figure 8. Pacific Decadal Oscillation (PDO) 2000 to 2013.............................................. 29
Figure 9. Pacific/North American (PNA) 2000 to 2013. .................................................. 30
Figure 10. North Atlantic Oscillation (NAO) 2000 to 2013. ............................................ 32
Figure 11. MACA Equal Area buffer. .............................................................................. 34
Figure 12. Selected Sites from Generalized Vegetation Classes at MACA. .................... 37
Figure 13. NDVI of MACA February 2000 to February 2013. ........................................ 38
Figure 14. NDVI of MACA buffer February 2000 to February 2013. ............................. 39
Figure 15. NDVI Yearly Aggregates of MACA and Park buffer 2000 to 2012. .............. 41
Figure 16. Total yearly precipitation in centimeters as measured by weather station at
Nolin River Lake 2000 to 2012. ....................................................................................... 42
Figure 17. Yearly mean, minimum, and maximum temperatures as measured by weather
station at Nolin River Lake 2000 to 2012. ........................................................................ 43
Figure 18. Summer mean, minimum, and maximum temperatures as measured by
weather station at Nolin River Lake 2000 to 2012. .......................................................... 43
Figure 19. Palmer Drought Severity Index for central Kentucky 2000 to 2012. .............. 44
vii
Figure 20. March 22nd NDVI Trend from 2000 to 2013................................................... 46
Figure 21. April 7th NDVI Trend from 2000 to 2013. ...................................................... 46
Figure 22. April 23rd NDVI Trend from 2000 to 2013. .................................................... 47
Figure 23. August 29th NDVI Trend from 2000 to 2012. ................................................. 48
Figure 24. September 14th NDVI Trend from 2000 to 2012. ........................................... 48
Figure 25. September 30th NDVI Trend from 2000 to 2012. ........................................... 49
Figure 26. October 16th NDVI Trend 2000 to 2012. ........................................................ 49
viii
LIST OF TABLES
Table 1. Nolin River Lake Weather Station Temperature and Precipitation Averages 2000
to 2012. ............................................................................................................................... 9
Table 2. Dominant Vegetation Classifications at MACA. ............................................... 15
Table 3. Multivariate ENSO Index Correlation with Temperature and Precipitation near
MACA............................................................................................................................... 28
Table 4. Pacific Decadal Oscillation Correlation with Temperature and Precipitation near
MACA............................................................................................................................... 29
Table 5. Pacific/North American Correlation with Temperature and Precipitation near
MACA............................................................................................................................... 30
Table 6. North Atlantic Oscillation Correlation with Temperature and Precipitation near
MACA............................................................................................................................... 32
Table 7. Multivariate ENSO Index Correlation with MACA summer NDVI. ................. 51
Table 8. Pacific Decadal Oscillation Correlation with MACA summer NDVI. .............. 51
Table 9. North Atlantic Oscillation Correlation with MACA summer NDVI. ................ 51
Table 10. Pacific/North American Correlation with MACA summer NDVI. .................. 51
Table 11. Multivariate ENSO Index Correlation with Vegetation Classes. ..................... 52
Table 12. Pacific Decadal Oscillation Correlation with Vegetation Classes. ................... 53
Table 13. North Atlantic Oscillation Correlation with Vegetation Classes. ..................... 54
Table 14. Pacific/North American Correlation with Vegetation Classes. ........................ 54
Table 15. MACA Summer NDVI Correlation with Temperature and Precipitation. ....... 58
ix
EASTERN DECIDUOUS FOREST PHENOLOGY AND VEGETATIVE VIGOR
TRENDS FROM 2000 TO 2013, MAMMOTH CAVE NATIONAL PARK, KY
Sean Taylor Hutchison
December 2013
76 Pages
Directed by: Dr. John All, Dr. Greg Goodrich, Dr. Jun Yan, and Dr. Jason Polk
Department of Geography and Geology
Western Kentucky University
Abstract
Global climate change is predicted to affect environmental systems at the midlatitudes, but the scope, severity, and outcomes of these impacts are yet to be fully
understood. This study focuses on the implications of short-term climate variability for
forests in central Kentucky. Using a Normalized Difference Vegetation Index (NDVI)
calculated from MODerate-Resolution Imaging Spectroradiometer (MODIS) instrument
data, the photosynthetic activity of vegetation at Mammoth Cave National Park (MACA)
is tracked from 2000 to 2013. Three methods were employed to examine the changes and
climate influences in vegetation over the study period: 1) aggregating the NDVI of the
Park by year and by summer months (June, July, and August) and examining how these
productivity trends could be influenced by precipitation and temperature fluctuations, 2)
examining the trend of the NDVI at selected dates throughout the study period to detect
phenological shifts around leaf-out and leaf-off, and 3) using a generalized vegetation
classification of MACA to clip the imagery based on areas of similar vegetation and then
testing correlations between those subsets and teleconnections. The results from the
aggregated NDVI show there is an insignificant negative trend. A negative relationship
between summer forest productivity at MACA and temperature was found, though more
data are needed to rigorously validate this result. Changes in phenology indicate forest
productivity is decreasing earlier each year throughout the study period. Finally, the
x
Multivariate ENSO Index and the Pacific Decadal Oscillation index are shown to have
significant positive correlations with the summer productivity of MACA during the study
period.
xi
Chapter 1: Introduction
As a result of the broad environmental movement of the last half-century, a new
ecosystem management paradigm has emerged that values the ecological function as well
as the economic use of resources. A wetland is now seen as a natural filtering facility for
water, where it was once seen as something to be drained for agriculture or development.
Forests are valued for their habitat and biodiversity and not just for their timber.
Resources must be examined for their function in the broader ecological context in order
to establish a sustainable model for resource management.
Currently, the greatest threats to these resources are human land use. The
seemingly insatiable need for “human place” has filled wetlands, cleared forests, and
leveled mountaintops. However, a new threat, anthropogenic climate change, will alter
the aesthetic, economic, and ecological value of resources around the globe (Houghton et
al. 2001). The effects of this trend are being witnessed in places like Glacier National
Park (Hall and Fagre 2003) and the glaciers at both poles. The effects of global warming
are particularly visible in the cryosphere because it has no adaptation mechanisms. The
effects on living systems, such as forests, can be more nuanced because of climate
adaptation and the large scales of these ecosystems (Thompson et al. 2009).
This research was designed to examine how mid-latitude forest productivity can
be influenced by short-term climate variability, principally fluctuations in temperature
and precipitation. The goal of this project was to create a baseline of information and
examine possible methodologies to examine how forest productivity reacts to short-term
variability, which may provide a better understanding of how long-term climate change
could impact similar forests at the mid-latitudes. This study employed satellite remote
1
sensing to monitor forest productivity from 2000 to 2013 at Mammoth Cave National
Park (MACA), located in Kentucky.
Three methods were used to study the changes in vegetation and the effects of
climate variability on the forests at MACA: 1) aggregating satellite-derived vegetation
productivity data by year and summer, and examining how these fluctuate in response to
temperature and precipitation changes, 2) examining forest productivity at selected earlyspring and late-summer dates for shifts in phenology, and 3) testing correlations between
teleconnections and broad vegetation groups within MACA.
Climate change presents an imminent threat to natural resources that demands a
comprehensive response. However, significant holes in the literature exist; specifically, in
regard to implications for mid-latitude locations such as Kentucky. Every species has
environmental conditions in which it optimally functions. Changes to either temperature
or precipitation can potentially have dramatic effects on a species, depending on its
ability to adapt. The role of a species in complex ecosystems can create codependent
relationships that many species rely on to survive. If one species is able to adapt to
changes in climate, but has a mutualistic relationship with another species that cannot
adapt, then negative outcomes for both species are possible (Kiers et al. 2010).
Understanding the response of vegetation to short-term climate variability is
critical to understanding the overall ecological response to longer-term changes. These
changes in climate can alter the growing season, structure, and composition of a forest
(Walther et al. 2002; Dale et al. 2009). In order to understand the extent of these impacts,
the physiological responses of tree species, community interaction within forest systems,
2
and rates of forest disturbances must be examined (Tylianakis et al. 2008; Dale et al.
2009; Allen et al. 2010; Gilman et al. 2010).
The broad spatial and temporal scale of the complex interaction between climate
and vegetation often limits the effectiveness of field experimentation (Aber et al. 2001).
Remote sensing has advantages over traditional field methods when studying regional
phenomena because of the larger scale of data collection. Remote sensing can be used to
create a Normalized Difference Vegetation Index (NDVI) over large areas. NDVI can
serve as a proxy measure for the amount of net primary productivity and biomass
detected in remotely sensed imagery (Birky 2001). Using this index, it is possible to
estimate the productivity and health of a forest and this measure was used to compare
with climate parameters during this research.
Although a considerable percent of the southern United States is forested, random
sampling is not practical due to the spatial distribution and density of forests and the
limitations of ground data collection methods. Additionally, disturbances such as fire and
insects can drastically alter the data and, without knowing the history of an area, errors
could result. National parks provide a suitable study area because they provide a large
area of forest with highly detailed and accurate management records that can be useful in
understanding past patterns in the data.
The National Park Service accounts for an estimated 1.3% of the total land in the
United States. A holding of such a large amount of land that is spread across a wide
geographic space will inevitably have valuable natural resources. Resource managers at
the park and network level are often aware of the threat of global climate change, but an
information gap exists concerning its potential effects. This research is designed with the
3
National Park Service’s need for climate change information in mind. One of the end
goals for this research is to design a methodology that can be easily exported to other
parks within the CUPN. Having such a methodology could be valuable to resource
managers as they attempt to mitigate the effects of climate change.
4
Chapter 2: Background
2.1 Climate Change Impacts on Forest Ecology
Researchers are confident that climate change will have a significant impact on
the productivity of forests (McNulty et al. 2001). However, global climate change will
have different impacts depending on the location. Zhao and Running (2010) found that
from 2000 to 2009, net primary productivity decreased in some regions, particularly in
the Southern Hemisphere, and increased in others such as the southeastern United States.
Climate change may influence forest composition across eastern United States (Iverson
and Prasad 2001) and the consequence of this could be an alteration in the function and
overall health of the broader ecosystem.
Although the global outcome of climate change is predicted to be a rise in average
temperature, the long-term regional and local effects remain largely unknown. According
to Cooter (1990), climate change is predicted to have extensive impacts on the vegetation
of the southern United States. Additionally, the tree farming industry, which is very
important economically for the southeastern United States, could experience altered
production rates (Irland et al. 2001; Dale et al. 2001; Burkett et al. 2001). Interruptions to
forest output could consequently have an impact on timber industries, which account
significantly in economies around the world. Negative impacts to forest output are not
certain, as some climate models suggest an increase in productivity; however, market and
social disruptions are expected regardless (Irland et al. 2001).
Even though a rise in global average temperature is a threat, it is a change in the
variability and extremes that may be most disruptive (Smith et al. 2005). The impacts
include changes to the species distribution, recovery rates, and overall health of a forest
5
system. Although species are capable of altering their range in response to climatic shifts,
mutualistic relationships and other biotic requirements complicate the adaptation process
(Hansen et al. 2001; Pearson and Dawson 2003).
Temperature, precipitation, and species dispersal have been examined in the
context of climate change, but according to Dale et al. (2000) the interaction between
forest disturbances and climate change remain an understudied yet important aspect for
research. Forest disturbances include: fires, hurricanes, tornados, ice storms, insects and
pathogens, exotic species, landslides, and drought. Each disturbance has varying
influence on physiological processes and species composition within a forest stand.
Climate change poses a threat to forests by potentially altering disturbance parameters
such as severity and frequency (Dale et al. 2001).
Fires are a major concern to resource managers, especially within the southeastern
United States where fires typically occur on an annual basis. However, Flannigan, Stocks,
and Wotton (2000) rebuke a belief that universal warming correlates with a universal rise
in fire, instead suggesting that fire is a result of complex interactions including
topography, weather, and fuel. Flannigan, Stocks, and Wotton demonstrate through
modeling that fires may only increase regionally where fire is already a concern.
Although drought is a simple concept to understand, there are different types of
drought with unique regional mechanisms. It is a challenge to model how climate change
will alter drought conditions without knowing the spatial context. Hanson and Weltzin
(2000) characterize the drought pattern of the eastern United States as random and
occasional. These types of droughts are typically limited in duration but are unpredictable
in terms of severity.
6
Potential impacts on forest productivity due to climate change have been
investigated in the southeastern region of the United States. Although few in number, the
old growth forests within the Cumberland region of Tennessee and Kentucky have
considerable diversity and high ecological integrity (McEwan et al. 2005). Land-use
changes pose a significant threat to the diversity of the Cumberland Forests by directly
altering forest stands and impacting local climate (Dale 2009). According to Burkett et al.
(2001) forest productivity within this region would shift northward over the next 100
years; as a result significant differences would develop between pine and hardwood
stands. Burkett et al. also suggests there is potential for an increase in fire impacts due to
a conversion of pine forest into pine savannah and grassland.
2.2 Study Area
The Cumberland-Piedmont Network (CUPN), part of the southeastern region of
the National Park Service (NPS), is a collection of fourteen parks in Tennessee,
Kentucky, Virginia, North Carolina, South Carolina, Georgia, and Alabama. The
common regional climate and species diversity of the area are well suited for this
research. Mammoth Cave National Park was the primary focus of this study because it
represents large portion (52,809 acres or 49%) of the land within the CUPN.
As stated in the Cumberland-Piedmont Network Vital Signs Monitoring Plan, it is
the mission of the National Park Service to preserve the state of the parks for the
enjoyment of future generations. Identifying climate change impacts is a high priority or
critical “vital sign” for which the NPS has mandated research (Leibfreid, Woodman, and
Thomas 2005). The research presented in this document could be valuable to resource
7
managers seeking to understand the possible impacts of climate variability on forest
systems.
Mammoth Cave National Park (MACA) is approximately 209.2km2 of protected
land located in central Kentucky (NPS 2013). About 96.5% of MACA is forested (Fry et
al. 2011), and the ratio of deciduous-to-evergreen was estimated to be approximately 70
to 30. The land within MACA has a history of use by humans for raw materials and
agriculture (Kennedy and Watson 1997), tourism (Algeo 2004), and habitat (Bailey 1933)
prior to its inclusion in the national park. The assorted historical uses of MACA have left
visible remnants of human influence on the landscape. Despite its history of exploitation,
MACA is a diverse ecological community that contains significant natural resources
(Leibfreid, Woodman, and Thomas 2005).
2.2.1 Climate
Hill and Mogil (2012) describe Kentucky as being at the intersection of cold air
from the north and warm air from the south, which creates a variety of weather patterns.
They also describe the temperature as nearly homogenous from the west end of the state
to the east (only a 2°C difference in the mean), and precipitation evenly distributed
throughout the year.
Weather data were collected for the study period through the National Climatic
Data Center (NCDC) from the Nolin River Lake Weather station, which is located
approximately a kilometer from the boundary of MACA. These summarized data, listed
in Table 1, shows the average temperature ranges from 1.6°C (35°F) in January to 25.4°C
8
(78°F) in July and August. The average high temperature ranges from 7.4°C (44°F) to
32.8°C (91°F), the average low temperature ranges from -4.4°C (24°F) to 18.4°C (65°F),
and the region receives approximately 136.7 cm (53.8 in) of rain per year (NCDC 2013a).
Historical weather data were also collected from the same weather station at Nolin
River Lake. The first full year of data available from that weather station is for 1965.
Comparisons between the historical record (1965 – 1999) and the study period record
(2000 – 2012) indicate some interesting differences. Analysis of the NCDC (2013a) data
show the study period has both higher measured temperatures and more precipitation
measured than the historical record. However, there are not enough data to make
definitive statements about warming and precipitation trends for the study period.
Table 1. Nolin River Lake Weather Station Temperature and Precipitation Averages 2000
to 2012.
Month Mean Temp (°C)
High Temp(°C)
Low Temp(°C)
Precip (cm)
Jan
1.6
7.4
-4.4
9.4
Feb
2.9
9.0
-3.5
9.3
Mar
8.5
15.4
1.5
10.1
Apr
13.9
21.4
6.4
14.2
May
19.2
26.2
12.2
16.7
Jun
23.7
30.9
16.3
9.6
Jul
25.4
32.4
18.4
12.2
Aug
25.4
32.8
18.0
8.7
Sep
21.5
28.7
14.2
9.6
Oct
14.4
21.8
7.0
9.6
Nov
8.9
15.9
2.0
11.7
Dec
2.8
8.5
-2.8
13.7
Data source: NCDC (2013a).
The rankings of months by temperature show that July, August, and June occupy
the top three rankings of average temperature in both periods. The rankings of months by
precipitation show that May was the wettest month on average in the historical and study
9
period record. August was the second driest month during the historical period and the
driest month in the study period.
2.2.2 Land Cover and Geology
The 2006 National Land Cover Dataset (Fry et al. 2011) was used to map land
cover at MACA and in the surrounding area. That dataset shows that MACA is almost
entirely forested; however, the area immediately outside of MACA is a mixture of forest,
agriculture/rangeland, and urban/built-up classes. In Figures 1 and 2, the dataset is
mapped at two spatial scales to show the details of the Park, as well as the broader region
that shows the non-contiguous forests surrounding MACA.
Mammoth Cave National Park is situated in a geologic landscape dominated by
carbonate rock known as karst. White et al. (1970, 88) writes, “the karst area of central
Kentucky is an outstanding example of one of the principal types of karst terrain in North
America.” The dissolution of these carbonate rocks over time can lead to the formation of
caves. This process is what forms limestone caves, such as Mammoth Cave (Palmer
1991), which is currently the longest cave in the world (Gulden 2013). Karstic geology
covers a vast majority of MACA. Figure 3 shows a generalized geologic map of MACA
created using data from the Kentucky Geological Survey (2002). The Chesterian age
rocks (so called for the geologic series in which they are grouped), and the Ste.
Genevieve and St. Louis formations are karstic rocks. These formations occupy
approximately 84% of MACA according to an analysis of the Kentucky Geological
Survey (2002) dataset. The Caseyville formation is non-karstic, and is only found in the
northern portion of MACA.
10
Figure 1. Mammoth Cave National Park Land Cover and Regional Context.
Map data source: 2006 National Land Cover Dataset (Fry et al. 2011).
11
Figure 2. Generalized Land Cover Classes Surrounding Mammoth Cave National Park.
Map data source: National Land Cover Dataset 2006 (Fry et al. 2011).
The Green River is the only major perennial surface stream at MACA. The karst
geology is highly porous and creates conduits that quickly drain surface waters. Soil
moisture, an important parameter for tree growth, is negatively impacted by the presence
of karst due to rapid infiltration of water to the subsurface. In periods of drought, this can
create especially difficult conditions for vegetation in karst areas where root depth is
constrained by shallow soils (Kukowski, Schwinning, and Schwartz 2011).
12
Figure 3. Generalized geologic map of Mammoth Cave National Park.
Map data sources: Kentucky Geological Survey (2002) and Palmer (1981).
13
2.2.3 Vegetation Classifications
Forest composition at MACA is mostly mixed, secondary growth with small
patches of old growth. The wide variation in elevation, slope, and aspect of MACA, with
its underlying karst geology and sprawling cave systems, have created many unique plant
communities. Smart and White (2010) note that the upland forests are dominated by oaks,
hickories, and other mixed hardwoods. Additionally, they report that the previously
cultivated fields, a remnant of the historic communities within the Park, and other
anthropogenically impacted areas are mostly dominated by eastern red cedar and Virginia
pine with some deciduous species around the perimeter.
The Center for Remote Sensing and Mapping Science (CRMS) at the University
of Georgia was contracted to create detailed vegetation maps of some of the national
parks within the Cumberland Piedmont network. The report by Jordan and Madden
(2010) used aerial photographs and Geographic Information Systems to delineate areas of
“relatively uniform vegetation” and assign classes according to the National Vegetation
Classification System as described in Jennings et al. (2006) and the Federal Geographic
Data Committee (2008). The aerial photograph used to delineate MACA was taken on
November 8th, 2001. Jordan and Madden (2010) found that “Interior Dry-Mesic White
Oak – Hickory Forest” was the most common vegetation classification at MACA, and it
accounted for approximately 21.25% of the area within the Park. Fifteen classifications
account for approximately 90% of the area within the Park (Table 2).
An assessment was performed at MACA to test the accuracy of the findings from
the Jordan and Madden (2010) report. Using a combination of field data, aerial
photography, and GIS, employees of NatureServe, the NPS, and volunteers evaluated the
14
accuracy by sampling evaluation points around the Park and comparing the CRMS map
to the findings on the ground. Smart and White (2010) report that the “accuracy
assessment” only found a 75.4% match with a Cohen’s kappa coefficient of (0.603).
Table 2. Dominant Vegetation Classifications at MACA.
Vegetation Classification
Interior Dry-Mesic White Oak – Hickory Forest
Eastern Red-cedar Successional Forest
Virginia Pine Successional Forest
Rich Appalachian Red Oak – Sugar Maple Forest
White Oak – Mixed Oak Dry-Mesic Alkaline Forest
Southeastern Successional Black Cherry Forest
Central Interior Beech – White Oak Forest
Beech – Maple Unglaciated Forest
Interior Low Plateau Chestnut Oak – Mixed Oak Forest
Southern Red Oak – Mixed Oak Forest
Virginia Pine – Red-cedar Successional Forest
Sycamore – Silver Maple Calcareous Floodplain Forest
Successional Tulip tree Forest (Circumneutral Type)
Shumard Oak – Chinquapin Oak Mesic Limestone Forest
Successional Tulip tree Forest (Acidic Type)
Total
Data source: Jordan and Madden (2010).
Hectares
4416
2222
1505
1286
1240
1206
1178
1172
1102
744
669
639
512
496
407
18794
% of Park
21.3%
10.7%
7.2%
6.2%
6.0%
5.8%
5.6%
5.6%
5.3%
3.6%
3.2%
3.0%
2.4%
2.3%
1.9%
90.1%
In order to account for the accuracy problems found in the CRMS report, this
study combined classes to produce three main vegetation types: Pine/Cedar, Oak, and
Mixed Hardwood. Pine and cedar species are common at MACA and have a similar
growing season. Oak species are also common, and the remainder of tree species can be
approximately classified as mixed hardwoods. Figure 4 shows how groups are distributed
throughout the Park. The mixed hardwoods group account for 51% of the area in MACA,
the Pine/Cedar group accounts for 22% and the Oak group accounts for 19%. The
remaining 8% of MACA not included in these three groups is primarily comprised of
human areas, the Green River and other water bodies, floodplain forests, and small
wetland marshes. Even with these groups the vegetation at MACA is highly
15
heterogeneous; this is likely due to the past human influences on the Park and the
secondary nature of the forest.
Figure 4. Generalized Vegetation Classes at MACA.
Map data source: modified from Jordan and Madden (2010).
16
Chapter 3: Geospatial Tools and Climate Indices
3.1 Remote Sensing
Forest disturbances require extensive monitoring that most traditional methods are
incapable of performing. Satellite remote sensing (RS) is well known for its ability to
examine the Earth at large spatial and temporal scales. Ecosystems can be monitored,
and depending on the resolution of the satellite, community-scale changes can also be
tracked. Biologists and resource managers are increasingly turning to RS as a data source
for ecological research (Kerr and Ostrovsky 2003).
The quality and applications of remotely sensed data are highly influenced by the
characteristics of the collection instrument and the post-processing techniques used to
prepare the data for consumption. Remote sensing can be broadly split into two types of
collection: active and passive. Active remote sensing emits energy and measures
reflectance of that energy. Underwater sonar, as used by submarines, is perhaps the most
recognizable example of active remote sensing. Passive collection measures the reflected
or emitted energy that is naturally occurring, chiefly from solar radiation. Further
references to the capabilities, strengths, or limitations of RS within this document apply
to passive collection systems.
There are many aspects of a remote sensing instrument; however, three things are
often discussed as the defining attributes: spatial, temporal, and spectral resolution.
Spatial resolution defines the minimum size of a ground-based object that can be
resolved. In other words, if a RS instrument has a spatial resolution of one meter, then all
objects one meter or bigger can be visually detected. It is not to be confused with spatial
coverage, or even the area of a single pixel within a RS image, which is defined by the
17
Instantaneous Field of View. Instruments with sub-meter spatial resolution are on the
high end of publically available imagery, though these datasets are typically only
available through purchase. Examples of satellites with very high spatial resolution
capabilities are GeoEye-1 (0.41m resolution) and WorldView-1 (0.5m resolution), both
privately operated. Most of the publically available RS data come from moderate to
lower-resolution instruments. Examples include the Thematic Mapper aboard Landsat 5
(30m to 120m resolution) and AVHRR aboard NOAA-10 (approximately 1.1km
resolution).
Temporal resolution defines the data collection interval. This is important because
some phenomena are short-lived and could occur between collection events. Many
satellites have a temporal resolution of one day, though post-processing operations can
redefine the temporal resolution to a longer interval. Spectral resolution defines the
ability of an instrument to distinguish between different parts of the electromagnetic
spectrum. An instrument with a high spectral resolution can differentiate between nearinfrared and mid-infrared light, whereas instruments with low spectral resolution might
only detect the presence of infrared reflectance without discrimination between different
wavelengths.
A major limitation for RS applications is that light must pass through the
atmosphere twice before it reaches the satellite and problems include moisture, aerosols,
and cloud cover, all of which can degrade the quality of data that are recorded. Collection
schemes and processing techniques are used to mitigate this loss of quality, but it is an
inherent flaw with RS that recorded data will never fully represent the ground truth.
Fieldwork and modeling must be used to validate the measurements (Turner et al. 2003).
18
Remote sensing provides information on vegetation and can directly detect
changes at large spatial scales, without the scaling problems associated with models
derived from ground observations (Cao et al. 2004). RS applications include estimating
agricultural production, determining species distribution of forests, and tracking the longterm changes in forest productivity. RS researchers have developed several vegetation
indices (VI) to mathematically estimate the amount of biomass. An index is created by
combining spectral bands in different ratios to form a value for each pixel.
3.1.1 Normalized Difference Vegetation Index
One of the most widely used VIs is the Normalized Difference Vegetation Index
(Campbell 2007). NDVI directly correlates with plant productivity. Monitoring
ecological performance and disturbance impacts is a primary use of this index (Pettorelli
et al. 2005). NDVI works by taking advantage of the predictable reflectance properties of
vegetation. Trees, shrubs, grasses, and any other plants that are live and green absorb
visible light and reflect others, particularly in the infrared region as part of the normal
photosynthetic process. NDVI is calculated by recording the reflectance of visible
wavelengths and the reflectance of a portion of infrared wavelengths nearest to the visible
spectrum, known as near-infrared or NIR. The NDVI formula is the difference between
the NIR and the visible reflectance over the sum of the NIR and the visible reflectance.
This ratio gives a value between -1 and 1. This ratio concept is illustrated in Figure 5.
As the visible light approaches or exceeds the amount of NIR reflectance, the
ratio will go down and approach -1. Lower numbers are typically indicative of barren or
rocky areas, and negative numbers typically indicate water, clouds, or snow. Conversely,
as NIR goes up in relation to visible reflectance, the ratio will approach 1. Higher
19
numbers are typically indicative of dense forests, such as those found at Mammoth Cave
National Park.
Figure 5. Normalized Difference Vegetation Index (NDVI) Diagram.
Figure source: Modified from NASA (2013).
NDVI has been used repeatedly to monitor productivity and measure the
influence of climate variability on tree growth and stress. Wang, Rich, and Price (2003)
used the NDVI and found that the vegetation productivity in their study area (Kansas)
during the growing season was positively correlated with current and previous
precipitation and negatively correlated with mean temperature during the middle of the
growing season. NDVI has been successfully used to research long-term trends in both
protected and non-protected areas throughout the world (see Pelkey, Stoner, and Caro
2000; Maselli 2004; Pouliot, Latifovic, and Olthof 2009; Tang et al. 2011). However,
there is a notable scarcity of studies using NDVI to examine long-term vegetation trends
20
in protected areas in Kentucky or any of the surrounding states. Hufkens et al. (2012)
used NDVI at MACA, but the study focused on the use of the satellite imagery with nearsurface measurements. The research presented here is the first known study to use the
NDVI to monitor inter-annual forest productivity trends at Mammoth Cave National
Park.
3.1.2 Data Acquisition and Processing
Many issues must be considered when choosing appropriate imagery for research.
Spatial, spectral, temporal, and cost factors are the most important that influence these
choices. Moderate Resolution Imaging Spectroradiometer (MODIS) was selected because
it met four critical criteria required for this study. The two most important criteria were
adequate spatial and spectral resolution. Typically, research into vegetation changes over
time requires at a minimum a regional scale spatial resolution. A fine spatial resolution
would provide too much data, and is not appropriate for land cover studies. Justice et al.
(1998) describes the Moderate Resolution Imaging Spectroradiometer (MODIS)
instrument as similar to the Advanced Very High Resolution Radiometer (AVHRR), but
with a higher spatial resolution and additional spectral bands. These bands are used to
produce image products that are well suited for studying vegetation with minimized
atmospheric interference.
The third criterion is a temporal resolution that provides data on at least a monthly
interval. The archive of available MODIS data begins in 2000 and spans to the present.
Observations are made every day, and a single image is generated from 16 days of
observations by compositing the imagery and computing the greatest value per raster cell
to achieve the maximum brightness, a technique called maximum value composite or
21
MVC (Huete, Justice, and van Leeuwen 1999). This acquisition scheme results in 22
images on average per year for each tile. The final criterion was cost-effectiveness.
Remotely sensed imagery can often be expensive, and high costs can limit the amount of
data a project can use. An advantage of MODIS is that it is free, so complete datasets can
easily be acquired.
NASA maintains many online warehouses of remotely sensed imagery. The Land
Processes (LP) Distributed Active Archive Network (DAAC), operated by the United
States Geological Survey (USGS), is home to the MODIS data that were used for this
study. LP DAAC provides direct FTP access to the data, which are filed according to
their tile, product type, and satellite platform (i.e. the spacecraft on which the MODIS
sensor is operating). Terra and Aqua are the only two spacecraft that use the MODIS
sensor at this time. Both spacecraft provide exactly the same spatial resolution, but the
temporal data acquisition is slightly staggered to provide increased coverage throughout a
year; therefore, a maximum of 44 images are possible per year.
Users do not typically access the raw MODIS data; instead, LP DAAC processes
the sensor data to create what are called data products. Each product contains multiple
layers of data that are best suited for different categories of remote sensing research.
There are products for studying Albedo, Fire, Temperature, Land Cover, and Vegetation
Indices. In total, there are 62 different products offered. The Vegetation Indices products
are best suited for this research. There are twelve total Vegetation Indices products, six
for each platform. The product called MOD13Q1 was used for data acquired from the
Terra platform, because it provides 250m resolutions and 16-day composites, whereas
22
other products are resampled to a lower resolution or monthly composites. MYD13Q1
was used for Aqua-based products, because it has the same specifications as MOD13Q1.
The VI products provided by LP DAAC are subjected to additional compositing
algorithms, in addition to the previously mentioned MVC. According to Huete et al.
(2002), this is necessary to correct for distortions introduced by off-nadir viewing angles
and the atmospheric corrections performed on the MODIS data. The constrained-view
angle-maximum value composite (CV-MVC) and the bidirectional reflectance
distribution function composite (BRDF) algorithms are therefore employed on all
MODIS-based VI products to select the best possible quality pixels for compositing.
MODIS uses a sinusoidal tiling system that divides the Earth into 595 equal area
squares or tiles. A tool, provided by the LP DAAC, converts latitude and longitude
coordinates into the tiling system for easier tile selection. The area for this study falls
entirely within the 11v5h tile, the extent of which is shown in Figure 6. The distortion of
the map seen in Figure 6 is due to the off-nadir viewing angle and sinusoidal projection
of the MODIS instrument.
The files have a naming scheme that retains useful information for the user. The
information is contained in segments separated by periods (‘.’). The first segment
indicates the product name, the second segment indicates the date of acquisition (in Julian
format), and the third segment indicates the tile within the sinusoidal system. Therefore
the example file ‘MOD13Q1.A2012241.h11v05.005.2012258032539.hdf’ indicates that
the product is MOD13Q1, the acquisition date is the 241st day of the year 2012, and the
data contained within the file is for tile h11v05.
23
Figure 6. Extent of Single MODIS tile.
Map data source: LP DAAC (2013).
The files are downloaded as HDF-EOS files, which is a format used by the Earth
Observing System operated by NASA to deliver layers of information as a single file.
Layer 1 of the HDF-EOS file is the Normalized Difference Vegetation Index (NDVI) and
24
is the most important layer for this study. An additional layer worth mentioning is layer
2, or the EVI layer. It is also a vegetation index, which seeks to increase sensitivity in
high biomass areas and reduce atmospheric interference. Other important layers - 4, 5, 6,
and 7 - are all reflectance bands representing red, NIR, blue, and MIR respectively. These
are used in the calculation of certain indices, among them the NDVI and EVI.
ArcGIS, a popular software package for managing, manipulating, and analyzing
geographic data, can subset the data and create a new image based on any of the original
layers. This step reduces the overall amount of data that must be managed, thus reducing
the computational requirements. Additionally, the new image was also converted from
the HDF-EOS format into the more compatible IMG format. ArcGIS was utilized to
subset images further by using a shapefile to clip the data to a selected area. The raster
was clipped to the exact extent and shape of the shapefile, leaving only the raster cells
that are within the area of a polygon. This research used several shapefiles, both created
and acquired, to clip the rasters to various boundaries.
ArcGIS provides many statistical measuring tools, of which the raster mean is the
most important for this study. The mean is calculated from the values found in the cells
of the image. The calculated mean of the raster corresponds to the average NDVI value
for that clipped image; the higher the average is, the more vegetation there is and viceversa.
The process (subset, clip, and calculate mean) required for analysis of a single
remotely sensed image had to be performed thousands of times for this research. This is
due to the number of images acquired for the study period and also the number of study
areas generated. In order to expedite this process, the Python programming language was
25
used to automate each step. The Python scripts for this automation process are described
in greater detail in Appendix A.
3.2 Teleconnections
The study of teleconnections examines how climate parameters interact over long
distances. Teleconnections can influence or modulate climate variables and thus can be
significantly related to other climate-driven disturbances such as wildfires (Dixon,
Goodrich, and Cooke 2008) and drought (Seager, Tzanova, and Nakamura 2009). This
section describes the teleconnections used in correlation analysis with the MACA NDVI
dataset. The index for each teleconnection is presented for the period 2000 to 2013.
Correlations between the monthly teleconnections indices and four weather
measurements (monthly mean, max, and minimum temperature and total monthly
precipitation) collected near MACA are also presented.
The four teleconnection indices employed are the Multivariate ENSO Index
(MEI), the Pacific Decadal Oscillation (PDO), the Pacific/North American (PNA), and
the North Atlantic Oscillation (NAO). All the data for these teleconnections were
retrieved from the Climate Prediction Center (CPC), part of the National Oceanic and
Atmospheric Administration (CPC 2013), except for the PDO data, which were retrieved
from the Joint Institute for the Study of the Atmosphere and Ocean at the University of
Washington (JISAO 2013).
3.2.1 Multivariate ENSO Index
One of the most widely recognized and studied teleconnections is the El Niño–
Southern Oscillation (ENSO). ENSO is a two-phase system where temperature and
26
surface pressure fluctuate between a warm and a cold phase, commonly known as El
Niño and La Niña respectively. Wolter and Timlin (1993, 2011) describe the multivariate
ENSO index as being calculated from six climate variables in the tropical Pacific: sealevel pressure, sea surface temperature, zonal (west-east direction) surface wind
component, meridional (north-south direction) surface wind component, surface air
temperature, and total cloudiness.
When MEI is positive, then ENSO is said to be in a positive, warm, or El Niño
phase. Conversely, when MEI is negative, ENSO is said to be in a negative, cold, or La
Niña phase. Wu, Hsieh, and Shabbar (2005) found that when ENSO is positive there is an
inverse effect on temperatures in the southeastern United States. Periods of significant
positive MEI from 2000 to 2013 occurred in 2002/2003, 2005, 2007, 2010, and 2012.
Periods of significant negative MEI from 2000 to 2013 occurred in 2000, 2008, and 2011.
The MEI values from 2000 to 2013 are plotted in Figure 7. There were no statistically
significant relationships between the MEI and concurrent weather measurements taken
near MACA (Table 3).
3.2.2 Pacific Decadal Oscillation
The Pacific Decadal Oscillation, or PDO, is also related to warming and cooling
trends in the Pacific like ENSO, but occurs on a much longer time scale. Mantua and
Hare (2002) describe the three main difference between PDO and ENSO as persistence
(PDO events can last 20 to 30 years while ENSO is typically 6 to 18 months), climatic
impacts (PDO influences extratropics with secondary effects on the tropics while the
opposite is true of ENSO), and drivers (ENSO mechanisms are well understood while the
PDO’s are less understood). Kurtzman and Scanlon (2006) found that PDO acts as a
27
modulator of ENSO in the Southeast, such that when PDO is positive it may enhance
ENSO, though these findings were for study areas in lower latitudes than MACA.
2013
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
-2.5
2000
Standardized Departure
Multivariate ENSO Index 2000 to 2013
Year
Figure 7. Multivariate ENSO Index (MEI) 2000 to 2013.
Data source: CPC (2013).
Table 3. Multivariate ENSO Index Correlation with Temperature and Precipitation near
MACA.
Mean
Max
Min
Total
Temp Temp Temp Precip
Multivariate ENSO
0.107
0.075
0.140
0.000
Pearson Correlation
Index
0.178
Sig. (2-tailed)
0.341
0.076
0.997
Results highlighted in grey are statistically significant.
Created using data from CPC (2013) and NCDC (2013a).
The variability of PDO from 2000 to 2013 was more positive in the first half of
the study period and more negative in the second half. The PDO values from 2000 to
2013 are plotted in Figure 8. There were two statistically significant relationships found
between PDO and weather measurements taken near MACA (Table 4). A negative
relationship exists between PDO and the mean and maximum temperature. As the PDO
index decreases the mean and maximum temperatures are increasing and vice-versa.
28
2013
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2.5
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
-2.5
-3
2000
Standardized Index
Pacific Decadal Oscillation 2000 to 2013
Year
Figure 8. Pacific Decadal Oscillation (PDO) 2000 to 2013.
Data source: JISAO (2013).
Table 4. Pacific Decadal Oscillation Correlation with Temperature and Precipitation near
MACA.
Mean
Max
Min
Total
Temp
Temp
Temp Precip
Pacific Decadal
-0.171
-0.201
-0.135
0.016
Pearson Correlation
Oscillation
0.031
Sig. (2-tailed)
0.011
0.087
0.839
Results highlighted in grey are statistically significant.
Created using data from JISAO (2013) and NCDC (2013a).
3.2.3 Pacific/North American
The Pacific/North American teleconnection, or PNA, describes a climate pattern
that relates midtropospheric circulation patterns of the Pacific Ocean and the North
American continent (Leathers, Yarnal, and Palecki 1991). Dixon, Goodrich, and Cooke
(2008) state that the positive phase of PNA is associated with below average
temperatures in the south-central and southeastern United States. Leathers and Palecki
(1992) found that winter and spring variance in mid-tropospheric flow over North
29
America is significantly related to PNA variations, whereas summer and autumn are not.
PNA exhibits a great amount of inter-annual variability.
The PNA index taken from 2000 to 2013 can be described as trending upward
from 2000 to the middle of the study period, then decreasing in the latter half. The PNA
values from 2000 to 2013 are plotted in Figure 9. There was a statistically significant,
negative relationship the PNA index and precipitation (Table 5). This indicates that as the
PNA index increases total precipitation is decreasing and vice-versa.
2013
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2.5
2
1.5
1
0.5
0
-0.5
-1
-1.5
-2
-2.5
-3
2000
Standardized Index
Pacific/North American 2000 to 2013
Year
Figure 9. Pacific/North American (PNA) 2000 to 2013.
Data source: CPC (2013).
Table 5. Pacific/North American Correlation with Temperature and Precipitation near
MACA.
Mean
Max
Min
Total
Temp
Temp
Temp Precip
Pacific/North
-0.073
-0.063
-0.082
-0.266
Pearson Correlation
American
0.358
Sig. (2-tailed)
0.430
Results highlighted in grey are statistically significant.
Created using data from (CPC 2013; NCDC 2013a).
30
0.300
0.001
3.2.4 North Atlantic Oscillation
The North Atlantic Oscillation, or NAO, index describes the surface pressure
difference between the higher latitudes of the North Atlantic near Iceland and the central
North Atlantic near the Azores (Rodwell, Rowell, and Folland 1999). This fluctuation
causes changes in surface westerlies in the Atlantic and in Europe (Hurrell 1995). When
the pattern is positive, the westerlies are strong, but in its negative phase the pressures in
the high latitudes of the North Atlantic can block westerlies in eastern North America,
which can influence significant snowstorms. (Wanner et al 2001; Seager et al. 2010). The
effects of the North Atlantic Oscillation are more pronounced during winter months
(Wallace and Gutzler 1981; Hurrell and van Loon 1997).
The NAO pattern can be described as having a flat trend from 2000 to 2006, a
negative trend from 2006 to 2011, and a positive trend from 2012 to 2013. The NAO
values from 2000 to 2013 are plotted in Figure 10. There was a statistically significant,
positive relationship between the NAO index and total precipitation measured near
MACA as shown in Table 6. This indicates that as the NAO index is increasing total
precipitation is also increasing.
31
North Atlantic Oscillation 2000 to 2013
Standardized Index
3
2
1
0
-1
-2
2013
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
-3
Year
Figure 10. North Atlantic Oscillation (NAO) 2000 to 2013.
Data source: CPC (2013).
Table 6. North Atlantic Oscillation Correlation with Temperature and Precipitation near
MACA.
Mean
Max
Min
Total
Temp
Temp
Temp Precip
North Atlantic
-0.144
-0.144
-0.141
0.204
Pearson Correlation
Oscillation
0.069
Sig. (2-tailed)
0.069
Results highlighted in grey are statistically significant.
Created using data from CPC (2013) and NCDC (2013a).
32
0.075
0.009
Chapter 4: Methods
The MODIS data were acquired from February 18th 2000 (first available date for
the MOD13Q1 product data) to July 28th 2013. There are many software packages used
to work with remotely sensed imagery. Some specific software packages are required to
perform advanced analysis techniques, such as unsupervised classifications or change
detections. However, only a simple set of tasks was required of the software; therefore, it
was determined that the built-in raster tools of ArcGIS would suffice. Several setup
procedures were performed with the ArcGIS tools before analysis could be conducted on
the data. The LP DAAC-acquired product came with 12 layers, but this study is only
concerned with the first layer.
The first step was to separate out the pertinent layer. After the separation of the
NDVI layer, each raster was clipped to the boundary of MACA and other intra-park and
extra-park boundaries. The average NDVI value was calculated for each clipped raster
and placed into a spreadsheet. The Python programming language was used to automate
these steps. The Python scripts used for this automation process are found in Appendix A.
Three methods of examining the data were utilized. The first method involved
aggregating the NDVI of the Park and an equal area buffer around the Park (Figure 11)
by year and summer (July, July, and August) and then examined the linear trend using a
‘best-fits’ method. This was done in part because de Beurs and Henebry (2004) found
that using a using linear regression on an entire time series of this type was inadequate
and did not produce reliable results due to statistical assumption violations. The method
employed by this research emulates the methods of Forkel et al. (2013). They found that
data aggregation by year or season was able to circumvent the problems associated with
33
the inter-annual cycle, with the caveat that different types of aggregation have different
applications. The purpose of including the buffer in this analysis was to see how the
protected vegetation of the Park compared to unmanaged vegetation in the surrounding
forests and fields.
Figure 11. MACA Equal Area buffer.
Map created using data from NPS (2013).
The second method looked at the phenology of the Park and how the growing
season may be shifting from 2000 to 2013. Gunderson et al. (2012) used experimentation
to show that an increase in temperature of 2°C resulted in a 4 to 9 day earlier leaf
emergence in four tree species located in eastern Tennessee. Studies have shown that
34
increasing trends in mean temperature have caused earlier blooms in some species
(Beaubien and Hamann 2011; Polgar and Primack 2011). In Kentucky, the growing
season begins around the middle of April. To test if a similar effect was occurring at
MACA, the NDVI value was evaluated at selected dates surrounding the leaf emergence
event. This same method was used for selected dates at the end of the growing season
(mid-August to late September) to examine how productivity was being impacted during
the driest part of the growing season. The third method looked at the statistical
relationship between teleconnections and the NDVI of MACA and the vegetation study
areas. The MACA buffer was also used in this analysis, but was not included in the
results because the correlations and significance levels were nearly identical to the
MACA results.
The MACA boundary was used to clip the entire series for the Park dataset. The
area immediately outside the Park was clipped to be approximately the same area as
MACA, shown in Figure 11. Twelve study sites, shown in Figure 12, were selected from
the generalized vegetation map derived from the CRMS study (Jordan and Madden
2010). The three generalized vegetation types that comprised over 90% of the area of
MACA were Pine/Cedar, Oak, and Mixed Hardwood. Four study sites were selected for
each of these groups respectively. The study sites were chosen based upon their extent
and dominant coverage of the raster cells by one vegetation group. Efforts were made to
limit the interference of prescribed burns on the study sites; however, four vegetation
sites are located in areas that were administered prescribed burns in 2004 (Oak1, Oak2,
and MixedHardwood3) and 2010 (PineCedar3).
35
The NDVI for each study area was grouped based on month by combining the
two 16-day composites that most overlapped with a given month. Since the NDVI time
series is not normally distributed, it cannot be used directly in a many common
parametric statistical analyses. The data were prepared so that only summer-month values
(June, July, and August) were used. This operation removed the bimodal distribution of
the data and resulted in thirty-nine data points for each variable. The corresponding
months for the selected teleconnections (see section 3.2 of this document) were also
chosen. After this preprocessing step, bivariate analysis was used to compare the summer
months of NDVI against the corresponding months of the selected teleconnections: MEI,
PDO, PNA, and POA.
36
Figure 12. Selected Sites from Generalized Vegetation Classes at MACA.
Map data source: modified from Jordan and Madden (2010).
37
Chapter 5: Results and Discussion
5.1 Aggregated NDVI
The NDVI of the whole park was plotted from February 2000 to February 2013
(Figure 13). Most NDVI values in this time series peak around 0.9 in the summer, and
decline to about 0.5 in the winter; however, the summer peaks for 2011 and 2012 are
noticeably lower (approximately 2.26% lower than the average peak NDVI between 2000
and 2010).
NDVI of MACA February 2000 to February 2013
NDVI
0.95
0.85
0.65
0.55
0.45
Year
Figure 13. NDVI of MACA February 2000 to February 2013.
Created using data from LP DAAC (2013).
38
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
0.35
2000
NDVI
0.75
The NDVI of the buffer was also plotted from February 2000 to February 2013
(Figure 14). There is more variability in the buffer highs and lows, which had an average
NDVI value of 0.82 in the summer and 0.48 in the winter. The year with the lowest
NDVI summer average was 2007, with 2008 and 2011 being second and third,
respectively. The most productive and least productive years between the Park and the
buffer are very similar, but do not match exactly.
NDVI of MACA buffer February 2000 to February 2013
NDVI
0.95
0.85
NDVI
0.75
0.65
0.55
0.45
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
0.35
Year
Figure 14. NDVI of MACA buffer February 2000 to February 2013.
Created using data from LP DAAC (2013).
Anomalous data points were identified in the time series of MACA and the buffer
if a value had a greater than 30% deviation from the two neighboring values. These
anomalies were given new values by averaging the two neighboring values. These
39
averaged values were only used as the input for Figures 13 and 14, and were not used in
any of the aggregation or correlation analysis discussed in this chapter. This was done
with four values at MACA for the dates 12/19/2000, 12/03/2002, 2/18/2005, and
2/2/2010 and with four values for the buffer for the dates 12/19/2000, 1/17/2003,
12/19/2004, and 2/2/2010. The likely cause of the extreme fluctuations between two data
points is extensive snow or cloud cover.
Typically, simple linear regression could be performed on a data set to determine
the overall trend; however, due to the statistical limitations on full NDVI datasets (de
Beurs and Henebry 2004) the yearly aggregates were calculated instead as recommended
by Forkel et al. (2013). These aggregates were calculated for the yearly average NDVI of
the Park (MACA), the average NDVI in the summer months (June, July, and August) for
the Park (MACA Summer), the yearly average NDVI for the Park buffer (Buffer), and
the average NDVI in the summer months for the Park buffer (Buffer Summer). These
aggregates are plotted in Figure 15. Note that there are no yearly aggregate values for the
year 2000, because the data record did not begin until February of that year.
One major drawback to using aggregated NDVI data is that the number of data
points is limited. Each series was transformed from 310 data points to only 12 or 13. This
limits the statistics that can be performed because of the small sample size. However,
qualitative descriptions can be made based on the visual assessment of the aggregates.
A weather station near Mammoth Cave National Park at Nolin River Lake has been in
operation since 1964. Among the many measurements recorded by the weather station
are total precipitation and mean, minimum, and maximum temperature. These weather
data from 2000 to 2012 were averaged by year and summer as shown in Figures 16, 17,
40
and 18. Additionally, the historical (1965-1999) averages for precipitation and
temperature are also plotted to easily compare how the means of the study period
compare to the historical period.
NDVI Yearly Aggregates of MACA and Park Buffer 2000 to
2012
MACA
MACA Summer
Buffer
Buffer Summer
0.95
0.9
NDVI
0.85
0.8
0.75
0.7
0.65
0.6
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Year
Figure 15. NDVI Yearly Aggregates of MACA and Park buffer 2000 to 2012.
Created using data from LP DAAC (2013).
These charts show that 2003 was a wet year while also being a cool year. In both
study areas, 2003 had the highest peak NDVI (approximately 1.4% higher than the
average for the Park, and 2.9% higher than the average for the buffer). In 2007, the total
precipitation was about 8.89 cm (3.5 in) below average and the temperature was the
second highest as measured by the mean and the average maximum. The NDVI of the
41
‘Buffer Summer’ was the lowest recorded (approximately 3.5% below the average),
while the NDVI of ‘MACA Summer’ was only a little above average (approximately
0.7% above).
Annual Precipitation as measured by Weather Station at
Nolin River Lake 2000 to 2012
Annual Precipitation
Historical Annual Average (1964-2012)
Summer Precipitation
Precipitation in centimeters
180.0
160.0
140.0
120.0
100.0
80.0
60.0
40.0
20.0
0.0
2000
2002
2004
2006
2008
2010
2012
Year
Figure 16. Total yearly precipitation in centimeters as measured by weather station at
Nolin River Lake 2000 to 2012.
Data source: NCDC (2013a).
42
Temperature in Deg C
Yearly Mean, Minimum, and Maximum Temperatures as
measured by Weather Station at Nolin River Lake 2000 to
2012
Mean Temperature
Historical Mean Temp (1964-2012)
Minimum Temperature
Maximum Temperature
25.0
20.0
15.0
10.0
5.0
2000
2002
2004
2006
Year
2008
2010
2012
Figure 17. Yearly mean, minimum, and maximum temperatures as measured by weather
station at Nolin River Lake 2000 to 2012.
Data source: NCDC (2013a).
Temperature in Deg C
Summer Mean, Minimum, and Maximum Temperatures as
measured by Weather Station at Nolin River Lake 2000 to
2012
Mean Temperature
Historical Mean Temp (1964-2012)
Minimum Temperature
Maximum Temperature
35
30
25
20
15
10
2000
2002
2004
2006
Year
2008
2010
2012
Figure 18. Summer mean, minimum, and maximum temperatures as measured by
weather station at Nolin River Lake 2000 to 2012.
Data source: NCDC (2013a).
43
The Palmer Drought Severity Index (PDSI), a widely used measure of drought
based on temperature and rainfall (Alley 1984), was also examined for relationships with
NDVI. The PDSI for the central Kentucky region was obtained from the NCDC (2013b).
The PDSI generally ranges from -6 to 6 with negative numbers representing drought
periods, 0 representing normal conditions, and positive numbers representing wet
periods. The data were aggregated for each year to remove the intra-annual variability to
focus on drought versus non-drought years. The PDSI for central Kentucky is plotted in
Figure 19, which shows that 2000, 2007, and 2012 were the most drought-affected
summers, while 2003, 2004 and 2011 were the wettest summers. There are similar
patterns reflected in the aggregated NDVI data (Figure 15).
Palmer Drought Severity Index (PDSI) 2000 to 2012
for Central Kentucky
PDSI
Summer PDSI
4
3
2
0
-1
-2
-3
Year
Figure 19. Palmer Drought Severity Index for central Kentucky 2000 to 2012.
Data source: NCDC (2013b).
44
2012
2011
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
-4
2000
Value
1
5.2 Phenology Shifts
The earliest indication of leaf bloom in the NDVI data set is April 7th. Around this
time the NDVI values begin to rapidly rise until they peak around June. Data from April
7th were selected for every year to check for changes in the growing season throughout
the study period. The surrounding dates, March 22nd and April 23rd, were also selected to
test if the NDVI remained relatively constant before and after the initial leaf out period.
The trend line was plotted for these days using the ‘best-fits’ method. The trend
lines for March 22nd (Figure 20) remained relatively flat (slope of 0.00008) with
moderate variation between years. The NDVI for April 7th is trending up over time (slope
of 0.0035). It is a slight upward trend over the study period, but the inter-annual
variability is considerably high (Figure 21). The trend line for April 23rd (Figure 22) was
flat (slope of 0.0002), and with a smaller, though considerable amount of inter-annual
variability.
The Coefficient of Variance (CV), a measure that is useful for comparing
variation between data with different means, was calculated for each of these dates. The
higher the CV, the more variation compared to other CV values. The CVs for March 22nd,
April 7th, and April 23rd are 0.102, 0.133, and 0.098 respectively. This indicates there is
the most variation within the April 7th time series, approximately when leaf bloom is
beginning to occur. The limitation of too few data points is a significant drawback, and
prevents linear trend analysis. Despite these issues, it is still useful to examine these
trends at individual dates from a qualitative perspective.
45
NDVI at March 22nd 2000 to 2013
NDVI
Trendline
0.9
NDVI Value
0.8
0.7
0.6
0.5
0.4
2000
2002
2004
2006
2008
2010
2012
Year
Figure 20. March 22nd NDVI Trend from 2000 to 2013.
Created using data from LP DAAC (2013).
NDVI at April 7th 2000 to 2013
NDVI
Trendline
0.9
NDVI Value
0.8
0.7
0.6
0.5
0.4
2000
2002
2004
2006
2008
2010
Year
Figure 21. April 7th NDVI Trend from 2000 to 2013.
Created using data from LP DAAC (2013).
46
2012
NDVI at April 23rd 2000 to 2013
NDVI
Trendline
0.9
NDVI Value
0.8
0.7
0.6
0.5
0.4
2000
2002
2004
2006
2008
2010
2012
Year
Figure 22. April 23rd NDVI Trend from 2000 to 2013.
Created using data from LP DAAC (2013).
The same process was repeated for the end of the growing season and beginning
of leaf abscission (leaf off). A decreasing trend is calculated from the NDVI values taken
on August 29th (Figure 23; slope of -0.005) and September 14th (Figure 24; also a slope of
-0.005) with a low amount of variance between the inter-annual values. A decreasing
trend is also found at the September 30th measurement (Figure 25; slope of -0.0034), but
with more variance between the inter-annual values. All other pre-winter measurements
after September 30th have large drops in the NDVI and increased variation similar to the
values shown in the trend for October 16th (Figure 26; slope of -0.0029).
47
NDVI at August 29th 2000 to 2012
NDVI
Trendline
0.9
NDVI Value
0.8
0.7
0.6
0.5
0.4
2000
2002
2004
2006
2008
2010
2012
Year
Figure 23. August 29th NDVI Trend from 2000 to 2012.
Created using data from LP DAAC (2013).
NDVI at September 14th 2000 to 2012
NDVI
Trendline
0.9
NDVI Value
0.8
0.7
0.6
0.5
0.4
2000
2002
2004
2006
2008
2010
Year
Figure 24. September 14th NDVI Trend from 2000 to 2012.
Created using data from LP DAAC (2013).
48
2012
NDVI at September 30th 2000 to 2012
NDVI
Trendline
0.9
NDVI Value
0.8
0.7
0.6
0.5
0.4
2000
2002
2004
2006
2008
2010
2012
Year
Figure 25. September 30th NDVI Trend from 2000 to 2012.
Created using data from LP DAAC (2013).
NDVI at October 16th 2000 to 2012
NDVI
Trendline
0.9
NDVI Value
0.8
0.7
0.6
0.5
0.4
2000
2002
2004
2006
2008
Year
Figure 26. October 16th NDVI Trend 2000 to 2012.
Created using data from LP DAAC (2013).
49
2010
2012
The CV values for August 29th, September 14th, September 30th, and October 16th
are 0.024, 0.024, 0.028, and 0.080 respectively. The low CV values for August 29th,
September 14th, and September 30th increase the probability that the negative trends
shown in their respective graphs are significant over time; however, it is not possible to
determine the significance currently due to the low number of data points.
5.3 NDVI and Teleconnections
The bivariate analysis for the selected summer NDVI values and the
teleconnections yielded mixed results. Table 3 through 6 shows the analysis for each of
the selected teleconnections compared to the summer NDVI of MACA. A one-month
time difference between the teleconnection and the NDVI values was also tested. This
test is either denoted as a ‘lag’ for NDVI readings that occur one month after the
teleconnection data, or as a ‘lead’ for teleconnection data as measured one month before
the NDVI values. Significant results in the tables are highlighted in grey.
The MEI had a significant, but moderate, relationship with the summer NDVI of
MACA. There was no significant relationship found between the MEI and the one-month
lags tested for MACA. The results of this analysis are shown in Table 7. The Pacific
Decadal Oscillation showed significant and strong relationships with the summer NDVI
for the Park. This was the most significant and strongest relationship found between the
NDVI of MACA and any teleconnection. Table 8 shows the results of this correlation
analysis. Only the summer NDVI for MACA at a one-month lag had a significant
relationship when the values were tested with the NAO teleconnection, though this
relationship can be described as moderate to strong with a Pearson’s r-value of 0.401.
These results are shown in Table 9. There were no significant relationships found
50
between the NDVI values of the Park and the Pacific/North American teleconnection as
shown in Table 10.
Table 7. Multivariate ENSO Index Correlation with MACA summer NDVI.
MACA One-Month
MACA
Lag
Multivariate ENSO
0.362
0.207
Pearson Correlation
Index
0.024
Sig. (2-tailed)
0.207
Results highlighted in grey are statistically significant.
Created using data from CPC (2013) and LP DAAC (2013).
Table 8. Pacific Decadal Oscillation Correlation with MACA summer NDVI.
MACA One-Month
MACA
Lag
Monthly Pacific
0.529
0.449
Decadal Oscillation Pearson Correlation
0.001
Sig. (2-tailed)
0.004
Results highlighted in grey are statistically significant.
Created using data from JISAO (2013) and LP DAAC (2013).
Table 9. North Atlantic Oscillation Correlation with MACA summer NDVI.
MACA One-Month
MACA
Lag
Monthly North
0.215
0.401
Atlantic Oscillation Pearson Correlation
0.188
Sig. (2-tailed)
0.011
Results highlighted in grey are statistically significant.
Created using data from CPC (2013) and LP DAAC (2013).
Table 10. Pacific/North American Correlation with MACA summer NDVI.
MACA One-Month
MACA
Lag
Monthly Pacific/
-0.204
-0.156
North American Pearson Correlation
0.213
Sig. (2-tailed)
0.343
Results highlighted in grey are statistically significant.
Created using data from CPC (2013) and LP DAAC (2013).
Tables 11 through 14 show the same analysis, but performed with the twelve
vegetation study areas clipped from the Park based on the Pine/Cedar, Oak, and Mixed
Hardwood groupings. The means of the each vegetation group were also calculated to
51
create vegetation group averages. This was done to detect if characteristics other than the
dominant type of vegetation (e.g. topography, geology, etc.) could be more influential on
the NDVI values than the behavior of the vegetation itself.
There were significant relationships between the MEI and the PineCedar3, Oak2,
MixedHardwood3, and MixedHardwood4, and the mixed hardwood average study areas
as shown in Table 11. No significant relationships were found between the study areas
and the MEI with a one-month lead time.
Table 11. Multivariate ENSO Index Correlation with Vegetation Classes.
Multivariate ENSO Index
MEI One-Month Lead
Pearson’s R
Significance
Pearson’s R
Significance
0.072
0.663
-0.111
0.500
PineCedar1
0.103
0.534
0.004
0.980
PineCedar2
0.359
0.025
0.211
0.198
PineCedar3
0.213
0.192
0.108
0.514
PineCedar4
0.223
0.153
0.059
0.723
PC Average*
0.240
0.141
0.124
0.450
Oak1
0.370
0.020
0.216
0.187
Oak2
0.207
0.207
0.061
0.714
Oak3
0.002
0.992
-0.138
0.402
Oak4
0.293
0.071
0.112
0.449
Oak Average*
0.176
0.284
0.084
0.610
MixedHardwood1
0.234
0.152
0.209
0.201
MixedHardwood2
0.370
0.020
0.178
0.279
MixedHardwood3
0.328
0.042
0.238
0.145
MixedHardwood4
0.380
0.017
0.235
0.149
MH Average*
* The average NDVI for the four vegetation study areas of its respective vegetation group.
Results highlighted in grey are statistically significant.
Created using data from CPC (2013), LP DAAC (2013), and Jordan and Madden (2010).
The PDO teleconnection had the most significant relationships with the vegetation
study areas, though there was not a particular class of vegetation that was more strongly
correlated than any other. PDO was shown earlier (Table 8) to have significant
relationships with the Park at concurrent and lag times. This is also true with PDO and
the vegetation classes. All the vegetation group averages were also found to be
52
significant. Only one study area (PineCedar1) had a significant relationship with the
concurrent PDO values and not the one-month lead time as shown in Table 12.
Table 12. Pacific Decadal Oscillation Correlation with Vegetation Classes.
Pacific Decadal Oscillation
PDO One-Month Lead
Pearson’s R Significance
Pearson’s R
Significance
0.357
0.026
0.279
0.086
PineCedar1
0.152
0.355
0.128
0.439
PineCedar2
0.498
0.001
0.473
0.002
PineCedar3
0.214
0.191
0.275
0.090
PineCedar4
0.391
0.014
0.366
0.022
PC Average*
0.143
0.385
0.162
0.324
Oak1
0.419
0.008
0.434
0.006
Oak2
0.514
0.001
0.406
0.010
Oak3
0.452
0.004
0.350
0.029
Oak4
0.434
0.006
0.400
0.012
Oak Average*
0.170
0.301
0.084
0.612
MixedHardwood1
0.407
0.010
0.343
0.033
MixedHardwood2
0.377
0.018
0.360
0.024
MixedHardwood3
0.413
0.009
0.339
0.035
MixedHardwood4
0.461
0.003
0.386
0.015
MH Average*
* The average NDVI for the four vegetation study areas of its respective vegetation group.
Results highlighted in grey are statistically significant.
Created using data from JISAO (2013), LP DAAC (2013), and Jordan and Madden
(2010).
The relationship between NAO and the Oak3 and Oak4 study areas were the only
significant relationships found with this teleconnection (Table 13). Only one significant
relationship was found between the vegetation classes and the PNA teleconnections. The
study area Oak4 had a significant, but negative relationship with the Pacific/North
American teleconnection at a one-month lead time (Table 14).
53
Table 13. North Atlantic Oscillation Correlation with Vegetation Classes.
North Atlantic Oscillation
NAO One-Month Lead
Pearson’s R Significance
Pearson’s R
Significance
0.18
0.273
-0.111
0.500
PineCedar1
0.059
0.721
0.004
0.980
PineCedar2
0.283
0.081
0.211
0.198
PineCedar3
0.103
0.533
0.108
0.514
PineCedar4
0.198
0.226
0.340
0.034
PC Average*
-0.021
0.899
0.124
0.450
Oak1
0.197
0.231
0.216
0.187
Oak2
0.399
0.012
0.061
0.714
Oak3
0.352
0.028
-0.138
0.402
Oak4
0.221
0.175
0.291
0.072
Oak Average*
0.236
0.148
0.084
0.610
MixedHardwood1
0.215
0.189
0.209
0.201
MixedHardwood2
0.112
0.497
0.178
0.279
MixedHardwood3
0.117
0.478
0.238
0.145
MixedHardwood4
0.228
0.162
0.389
0.015
MH Average*
* The average NDVI for the four vegetation study areas of its respective vegetation group.
Results highlighted in grey are statistically significant.
Created using data from CPC (2013), LP DAAC (2013), and Jordan and Madden (2010).
Table 14. Pacific/North American Correlation with Vegetation Classes.
Pacific/North American
PNA One-Month Lead
Pearson’s R
Significance
Pearson’s R
Significance
-0.131
0.428
-0.233
0.154
PineCedar1
-0.003
0.984
-0.287
0.076
PineCedar2
-0.106
0.522
-0.175
0.286
PineCedar3
-0.202
0.217
-0.154
0.350
PineCedar4
-0.138
0.403
-0.284
0.080
PC Average*
0.013
0.939
-0.049
0.768
Oak1
-0.118
0.474
-0.205
0.211
Oak2
-0.110
0.503
-0.156
0.344
Oak3
-0.152
0.354
-0.348
0.030
Oak4
-0.090
0.584
-0.214
0.191
Oak Average*
-0.285
0.078
-0.099
0.551
MixedHardwood1
-0.303
0.060
0.021
0.897
MixedHardwood2
-0.131
0.427
-0.173
0.293
MixedHardwood3
-0.205
0.210
-0.131
0.427
MixedHardwood4
-0.305
0.059
-0.133
0.420
MH Average*
* The average NDVI for the four vegetation study areas of its respective vegetation group.
Results highlighted in grey are statistically significant.
Created using data from CPC (2013), LP DAAC (2013), and Jordan and Madden (2010).
54
5.4 Data Interpretation
Several values in the aggregate and phenology analyses require additional
investigation, either because they are anomalous or significant weather-related
disturbances (e.g. snow, ice, and flooding) or periods of extreme weather (i.e. drought)
occurred at the time the data were acquired. The relationship between forest productivity
and these significant weather events/periods is important to understand because
temperature and precipitation are critical parameters of plant growth, and disturbances
can severely damage the health and productivity of vegetation. Understanding this
relationship may explain some of the inter-annual variation in the aggregate NDVI, and
could serve as a baseline of information about how the long-term climate may impact the
forests at MACA.
The year 2000 was a significant drought period for the region. According to the
National Climatic Data Center (2013c), 2000 was the 20th driest June to August period
using the Palmer Drought Severity Index for Climate Division 2 in Kentucky (where
MACA is located) in its 119-year record. The PDSI value for the period in 2000 is -2.14.
Between 2000 and 2003 the PDSI went from -2.14 to 2.72 (the 9th wettest June to August
period in the PDSI record for this region). This transition from dry to wet conditions is
reflected in the summer NDVI data. The summer NDVI steadily increased from 2000 to
2003 (see Figure 15). This is an expected relationship because it is well established that
periods of droughts, especially in summer months, can cause leaf stress that would lower
the NDVI. As precipitation increased and mean and maximum temperatures decreased,
the summer vegetation at MACA appeared to respond by being more productive.
55
The summer NDVI for the buffer around MACA dropped significantly in 2007;
however, the summer NDVI at MACA or the yearly NDVI aggregates either increased or
were flat compared to 2006, and the average summer NDVI for the entire study period.
The summer PDSI for the area was -1.90, the 21st driest June to August period on record;
however, the January to December PDSI was considerably lower (-0.74), indicating that
the drought was primarily concentrated in the summer that year (NCDC 2013c). The
summer mean and maximum temperatures were above their historical averages (ranked
the 20th warmest June to August period in the NCDC record), and the yearly temperature
averages were also warmer than their respective historical average. There was a large
difference in precipitation that year between the summer and yearly averages. Summer
precipitation was at its lowest recorded value during the study period, but the yearly
average was near both the study period and historical averages. A difference in land
cover is the most likely explanation of the observed difference between the buffer and
MACA NDVI. The buffer contains agricultural lands, while MACA is almost entirely
forested (see Figure 1). During periods of intense drought, cropland may be more prone
to heat stress than dense, robust forests such as those found at MACA. According to
Goodrich et al. (2011) and Craft (2011), the quality and yield of crops in Kentucky,
particularly corn, were impacted by the drought of 2007.
On March 22nd, 2007 the NDVI value record for the Park was approximately 30%
above the average of other March 22nd NDVI data collected from 2000 to 2013 (shown in
Figure 20). However, by the next available date (April 7th), the NDVI value of the park
was 8% below the average (shown in Figure 21). This is likely due to unseasonably warm
temperatures in March, and a frost event and unseasonably cold temperatures in April.
56
March of 2007 was the fifth warmest March in the NCDC (2013c) record for Climate
Division 2 of Kentucky. These warm conditions encouraged early spikes in vegetation
productivity. However, in early April, a freeze event caused widespread damage over
wide swaths of the eastern United States (Gu et al. 2008; Warmund, Guinan, and
Fernandez 2008), which is seen in the NDVI data. The damage to the productivity of the
Park was visible through at least the beginning of May, which had the lowest recorded
NDVI value throughout the study period.
A similar spike above the average is also seen in 2012. This is likely due to the
unseasonably warm temperatures during that period. March of 2012 is currently the
warmest March in the NCDC (2013c) record. There was no damaging frost that year; in
fact, both April and May were unseasonably warm as well. The NDVI values for March
22nd, April 7th and April 23rd 2012 were 18%, 30%, and 9% above their study period
averages, respectively.
In January 2009, an extensive ice storm impacted the central Kentucky region,
including the Mammoth Cave area. The ice storm caused widespread damage to trees
through loss of limbs and uprooting (Vowels 2012). The NDVI value around March 22nd
2009 (about two months after the ice storm occurred) was average compared to the entire
study period. However, the April 7th and April 23rd 2009 values were well below their
averages (21% and 26% respectively). This is likely due to the extensive damage to the
trees that created a lasting, negative impact on MACA in 2009.
The calculated slope of the best-fit trend lines for the phenology analysis indicates
considerable changes to the productivity over the study period. The difference between
winter and summer NDVI is on average approximately 0.4. If the trends for August 29th
57
and September 14th continue, then the NDVI in late summer would decrease by 0.1 in just
20 years starting in 2000. The implication would be that MACA is decreasing in
productivity earlier in the year. Possible explanation for this trend is a shift in the
growing season or increased heat stress due to rises in temperature or the occurrence of
drought. Without more years of data, it is not currently possible to determine the
significance of this trend.
Although precipitation is a critical parameter for plant growth, analysis of the
aggregate NDVI values, especially for the summer months, indicates that temperature
may actually exert the more of an influence on forest productivity. Table 15 presents the
results of a correlation analysis between the ‘MACA Summer NDVI’ (defined as the
average NDVI of June, July, and August) for 2000 to 2012 and selected weather variables
gathered from the Nolin River Lake weather station (NCDC 2013a). The weather
variables are mean, maximum, and minimum temperature and total precipitation. These
values were calculated as the averages of the summer months (June, July, and August) for
temperature and the total for precipitation in those same months.
Table 15. MACA Summer NDVI Correlation with Temperature and Precipitation.
Mean
Max
Min
Total
Temp
Temp
Temp
Precip
MACA
Pearson
-0.683
-0.682
-0.479
0.274
Summer NDVI Correlation
0.010
0.010
0.097
0.365
Sig. (2-tailed)
Results highlighted in grey are statistically significant.
Created using data from LP DAAC (2013) and CPC (2013).
The analysis reveals that there are significant, negative relationships between the
MACA summer NDVI and the mean and maximum summer temperatures. Although
these are interesting results, they are limited because of the small sample size (n=12) of
the analysis. Despite the relatively low number of data points, these results are an
58
indication that temperature exerts significant control over vegetation productivity at
MACA. This is likely manifested as heat stress during moderate to intense drought
periods, though more data is needed to make any definitive conclusions.
If mean and maximum temperature indeed exert the most influence over NDVI of
all the variables examined, then it is not surprising that the Pacific Decadal Oscillation is
the most significant correlation found between the selected teleconnections and the
vegetation of MACA. PDO had significant relationships with the summer mean and
maximum temperature, but the other teleconnections did not. ENSO, PNA, and NAO
may have more influence over winter weather, so future analysis could focus on that
relationship. However, it may be hard to correlate winter weather with NDVI at MACA,
because about 70% of the trees at MACA are deciduous.
Prescribed burns can have a lasting impact on the growth and productivity of
vegetation. Depending on the intensity and the spatial extent of the fire, the NDVI data
possibly could be affected. However, the burns in the Park are usually of very low
intensity and there is no indication from the NDVI data that the use of prescribed burns
near the vegetation study sites significantly influenced the results. Additional work is
needed to better understand how prescribed burns are used at MACA, and how their use
could impact the short and long-term productivity of the vegetation.
59
Chapter 6: Conclusions
The yearly and seasonal aggregation of NDVI data is both useful and limiting.
Positive and negative trends are visible in the data, and seem to be influenced
significantly by mean and maximum temperature and drought events. Significant
deviations from the mean of annual and summer precipitation are also reflected in the
aggregated NDVI data. However, the low number of data points in the aggregated data
set limits the statistical tools available to reliably test the correlations between the NDVI
values and weather variables.
The equal area buffer provides a useful check of NDVI values at the Park
compared to other vegetation in the area. The buffer is also useful to test if the
differences between managed and unmanaged vegetation were influential on the
productivity. The NDVI values of the Park were comparable to those in the buffer,
although the NDVI values of the Park were consistently higher throughout the year.
Conclusions about managed and unmanaged areas are difficult to make based on the
aggregated data alone, but it is clear that the NDVI values in the unmanaged buffer had a
higher variability compared to the Park. The higher variability in the buffer time series is
likely due to the human disturbances, such as varied agricultural uses of the area.
An earlier growing season is suggested but not demonstrated based on ‘leaf out’
NDVI dates (March 22, April 7, and April 23). However, based on the late growing
season dates (August 29, September 14, and September 30), it is likely that forest
productivity is decreasing earlier in the season each year. Late summer season decreases
in NDVI over time are possibly related to increases in maximum temperature. The
relatively small data set currently limits the significance of these results.
60
A strong, positive relationship was found between the Park and the PDO
teleconnection. PDO was the strongest correlation found between the NDVI of the Park
and any of the teleconnections tested. PDO was significantly correlated with mean and
maximum temperature during the summer, but not with minimum temperature or
precipitation. A significant, positive relationship between ENSO and NDVI was also
found; however, the relationship was not as strong or significant as with PDO. ENSO was
correlated with summer temperature or precipitation. This was also true of the NAO and
PNA teleconnections. The influence of ENSO, NAO, and PNA over North American
winter weather patterns rather than summer may explain this result.
The correlations between vegetation group productivity and the teleconnections
were similar in significance to the results for the Park as a whole. PDO was found to
have the strongest correlation with all the different vegetation groups, while ENSO,
NAO, and PNA were less significantly correlated with fewer groups. The variation in the
correlations between similar vegetation study areas is interpreted to mean that other
factors, such as topography and the lack of homogeneity may have a strong influence on
the NDVI values. Future studies that use this technique would benefit from using
homogenous tree stands for study areas if they could be found.
The impacts of long-term climate change were too subtle to detect during the
short time period of this study. Mammoth Cave National Park has robust forests that are
capable of adapting to climate stressors and natural disturbances. More research is needed
at the ground level to determine the physical responses by species and community type
before broad conclusions can be drawn about the impacts of long-term climate change on
the forests of Mammoth Cave. However, short-term climate variability, such as
61
fluctuations in temperature and periods of drought is shown to have a significant impact
on the productivity of the forests at MACA. Mean and maximum temperatures are
indicated as the biggest influences on the productivity of the forests over the study
period.
Future studies should focus on interpreting the non-aggregated time series to
statistically determine if there is a long-term trend in the NDVI. The use of nonparametric statistics and seasonal modeling to adjust for the cyclical nature of the data are
likely needed to determine a trend from the complete NDVI dataset. The non-parametric
Mann-Kendall trend test is suggested by de Beurs and Henebry (2004), because it is able
to avoid the problems associated with temporal autocorrelation, or inter-annual cycles, in
the data. Determining the long-term trend on a per pixel basis, rather than averaging the
entire Park, may also be useful for future research. Such a technique could determine if
increases and decreases in NDVI are spatially autocorrelated. Finally, examining how
other large forests in the region, such as Daniel Boone National Forest, behaved over the
same period could give further insights into the role of geography on climate change
impacts.
62
APPENDIX A: RASTER PROCESSING SCRIPTS
During the data processing of this research, it became necessary to perform
repetitive processing tasks with the Remote Sensing data. Each HDF file was first
subsetted to produce the NDVI file, then the NDVI file was clipped to the boundary of
the study area, and finally the mean value of the clipped raster was extracted and put into
a spreadsheet along with the corresponding Julian date. The dataset required for this
research necessitated this be done over 7500 times. Three scripts were written in the
Python programming language in order to expedite the subset, clipping, and recording
steps. Common Python operations were combined with functions from the ArcPy
package, which come standard with ArcGIS Desktop 10.0 and higher, to iterate over all
the remote sensing data and extract the required information.
Anyone who needs to process a large set of remote sensing data could benefit
from these scripts. The setup time for these scripts is minimal, assuming one already has
ArcGIS Desktop 10.0 or higher installed, and specific knowledge of Python is not
required. The scripts can be used with the IDLE editor included with ArcGIS. A standard
installation of the latest version of ArcGIS (version 10.2 at the time of this writing) will
place this at C:\Python27\ArcGIS10.2\Lib\idlelib\idle.pyw. Open this IDLE editor file
then, within the Python IDLE window, choose File, Open, and then navigate to the
location of the python script you wish to execute. Once the variables are set (i.e. folder
location for the data, folder location to output results, path to the shapefile used for
clipping, etc.), save the file then select Run and Run Module. A window will appear and
display the results of each operation as it iterates over the set of rasters.
63
The source code for the three scripts is included below. Copy the text of the script
to a blank text file and save the file with a .py extension. An important thing to remember
about Python is that the spaces or indentions highly influence the execution of the script.
Do not attempt to modify the hierarchy of the indentions.
The three scripts included below are:
1. subset.py: this script is used to iterate over a set of HDF files and save the desired
layer to a specific path. Additionally this script renames the new image using a
segment of the original file name. For this research it was desirable to maintain
only the Julian date of which the image is associated. The original file names are
formatted like this: MOD13Q1.A2012001.h11v05.005.2012019110526.hdf. Each
segment is separated by a period. The Julian date is located in the second
segment, so this script captures that segment, removes the first letter and saves the
new image as 2012001.img.
2. clip.py: this script iterates over a set of rasters and clips each using a specified
shapefile. An important variable to change in the arcpy.Clip_management
function is the second parameter, which is the extent of the shapefile. This
information can be found in ArcGIS within the properties of the shapefile on the
source tab. The order of the extents is left, bottom, right, and top with each value
separated by a space.
3. rastermean.py: this script iterates over a set of rasters, calculates the mean value
of all the cells within the raster and writes the output to a text file using the rasters
name as an identifier. Ideally the raster name will be the Julian date that was set
earlier by subset.py. This file can then be imported in excel or other spreadsheet
64
software easily as it is formatted as a comma separated value (CSV) file. The file
will be located in the same directory as the rasters you ran the script on.
These scripts are provided “as-is,” feel free to modify them however is necessary.
There are no licensing restrictions other than those required by Python and ESRI.
65
#subset.py
import os
import shutil
import arcpy
workspace = "C:\\Path\\To\\Rasters\\"
clippingSHP = "C:\\Path\\To\\boundary.shp"
savelocation = "C:\\Path\\To\\Output\\Location\\"
def shortenName(fullname):
namearr = fullname.split('.')
namer = namearr[1]
return namer[1:]
for file_ in os.listdir(workspace):
file_name, file_Extension = os.path.splitext(file_)
file_Extension = file_Extension.lower()
hdf = workspace+file_
if not os.path.isdir(hdf):
if file_Extension==".hdf":
outputname = shortenName(file_name)
print "Subsetting:"+outputimg
outputimg = savelocation+outputname+".img"
arcpy.ExtractSubDataset_management(hdf, outputimg, "0")
print "Done with: "+outputname
66
#clip.py
import os
import shutil
import arcpy
workspace = "C:\\Path\\To\\Rasters\\"
clippingSHP = "C:\\Path\\To\\boundary.shp"
savelocation = "C:\\Path\\To\\Output\\Location\\"
for file_ in os.listdir(workspace):
file_name, file_Extension = os.path.splitext(file_)
file_Extension = file_Extension.lower()
original = workspace+file_
if not os.path.isdir(workspace+file_):
if file_Extension==".img":
imgname = file_name
print "Clipping:"+imgname
img = savelocation+imgname+".img"
arcpy.Clip_management(original, "-7639791.896018
4127825.340617 -7612579.323959 4144172.309948", img, clippingSHP, "",
"ClippingGeometry", "NO_MAINTAIN_EXTENT")
print "Done with: "+img
67
#rastermean.py
import arcpy
import glob
import os
os.chdir("C:\\Path\\To\\Rasters\\")
outputFile = open('RasterMeanOutput.txt','w')
for file in glob.glob("*.img"):
myval = arcpy.GetRasterProperties_management(file, "MEAN")
imgname = file[:-4]
outputFile.write(imgname+','+myval.getOutput(0)+'\n')
print "Done with: "+imgname+","+myval.getOutput(0)
outputFile.close()
68
Literature Cited
Aber J., R. P. Neilson, S. McNulty, J. M. Lenihan, D. Bachelet, and R. J. Drapek. 2001. Forest
Processes and Global Environmental Change: Predicting the Effects of Individual and
Multiple Stressors. BioScience 51(9): 735-751.
Algeo, K. 2004. Mammoth Cave and the making of place. Southeastern Geographer 44(1): 2747.
Allen, C. D., A. K. Macalady, H. Chenchouni, D. Bachelet, N. McDowell, M. Vennetier, T.
Kitzberger, A. Rigling, D. D. Breshears, E. H. Hogg, P. Gonzalez, R. Fensham, Z. Zhang,
J. Castro, N. Demidova, J-H. Lim, G. Allard, S.W. Running, A. Semerci, and N. Cobb.
2010. A global overview of drought and heat-induced tree mortality reveals emerging
climate change risks for forests. Forest Ecology and Management 259(4): 660-684.
Alley, W. M. 1984. The Palmer Drought Severity Index: Limitations and Assumptions. Journal
of Climate and Applied Meteorology 23(7): 1100-1109.
Bailey, V. 1933. Cave life of Kentucky. Mainly in the Mammoth Cave region. American
Midland Naturalist 14(5): 385-635.
Beaubien, E., and A. Hamann. 2011. Spring flowering response to climate change between 1936
and 2006 in Alberta, Canada. Bioscience 61(7): 514-524.
Birky, A. K. 2001. NDVI and a simple model of deciduous forest seasonal dynamics. Ecological
Modelling 143(1): 43-58.
Burkett, V., R. Ritschard, S. McNulty, J. J. O’Brien, R. Abt, J. Jones, U. Hatch, B. Murray, S.
Jagtap, and J. Cruise. 2001. Potential consequences of climate variability and change for
the Southeastern United States. In Climate Change Impacts in the United States:
Potential Consequences of Climate Change and Variability and Change., eds. J. M.
Melillo, A. C. Janetos, T. R. Karl, E. J. Barron, V. Burkett, T. E. Cecich, R. Corell, K.
Jacobs, L. Joyce, B. Miller, M. G. Morgan, E. A. Parson, R. G. Richels, and D. S.
Schimel, 137-164. Cambridge, UK: Cambridge University Press.
Campbell, J. B. 2007. Introduction to Remote Sensing 4th edition. New York: The Guilford Press.
Cao, M., S. D. Prince, J. Small, and S. J. Goetz. 2004. Remotely Sensed Interannual Variations
and Trends in Terrestrial Net Primary Productivity 1981-2000. Ecosystems 7(3): 233-242.
69
CPC (Climate Prediction Center). 2013. Climate and Weather Linkage. National Oceanic and
Atmospheric Administration. Available at
http://www.cpc.ncep.noaa.gov/products/precip/CWlink/MJO/climwx.shtml (last accessed
6 November 2013).
Cooter, E. J. 1990. The impact of climate change on continuous corn production in the Southern
U.S.A. Climatic Change 16(1): 53-82.
Craft, K. E. 2011. Economic Impacts of Drought on Kentucky Corn, Hay, and Soybeans.
Geoscience Master’s thesis. Department of Geography and Geology, Western Kentucky
University.
Dale, V. H., L. A. Joyce, S. McNulty, and R. P. Neilson. 2000. The interplay between climate
change, forests, and disturbances. The Science of the Total Environment 262(3): 201-204.
Dale, V. H., L. A. Joyce, S. McNulty, R. P. Neilson, M. P. Ayres, M. D. Flannigan, P. J. Hanson,
L. C. Irland, A. E. Lugo, C. J. Peterson, D. Simberloff, F. J. Swanson, B. J. Stocks, and
B. M. Wotton. 2001. Climate Change and Forest Disturbances. BioScience 51(9): 723734.
Dale, V. H., K. O. Lannom, M. L. Tharp, D. G. Hodges, and J. Fogel. 2009. Effects of Climate
Change, Land-use Change, and Invasive Species on the Ecology of the Cumberland
Forests. Canadian Journal of Forestry 39(2): 467-480.
de Beurs, K. M., and G. M. Henebry. 2004. Trend analysis of the Pathfinder AVHRR Land
(PAL) NDVI data for the deserts of Central Asia. Geoscience and Remote Sensing
Letters, IEEE 1(4): 282-286.
Dixon, P. G., G. B. Goodrich, and W. H. Cooke. 2008. Using Teleconnections to Predict
Wildfires in Mississippi. Monthly Weather Review 136(7): 2804 - 2811.
Federal Geographic Data Committee. 2008. National Vegetation Classification Standard,
Version 2 FGDC-STD-005-2008 (version 2). Reston, VA: Vegetation Subcommittee,
Federal Geographic Data Committee, FGDC Secretariat, U.S. Geological Survey.
Flannigan, M. D., B. J. Stocks, and B. M. Wotton. 2000. Climate change and forest fires. Science
of the Total Environment 262(3): 221-229.
Forkel, M., N. Carvalhais, J. Verbesselt, M. D. Mahecha, C. S. R. Neigh, and M. Reichstein.
2013. Trend Change Detection in NDVI Time Series: Effects of Inter-Annual Variability
and Methodology. Remote Sensing 5(5): 2113-2144.
70
Fry, J., G. Xian, S. Jin, J. Dewitt, C. Homer, L. Yang, C. Barnes, N. Herold, and J. Wickham.
2011. Completion of the 2006 National Land Cover Database for the Conterminous
United States, Photogrammetric Engineering and Remote Sensing 77(9): 858-864.
Gilman, S. E., M. C. Urban, J. Tewksbury, G. W. Gilchrist, and R. D. Holt. 2010. A framework
for community interactions under climate change. Trends in Ecology and Evolution
25(6): 325–331.
Goodrich, G. B., J. K. Thompson, S. D. Wingard, and K. J. Batson. 2011. The 2007 Mid-South
Summer Drought and Heat Wave in Historical Perspective. Southeastern
Geographer 51(3): 411-421.
Gu, L., P. J. Hanson, W. M. Post, D. P. Kaiser, B. Yang, R. Nemani, S. G. Pallardy, and T.
Meyers. 2008. The 2007 Eastern US spring freeze: increased cold damage in a warming
world?. BioScience 58(3): 253-262.
Gulden, B. 2013. Worlds Longest Caves. NSS Geo2 Committee On Long and Deep Caves.
Available at http://www.caverbob.com/wlong.htm (last accessed 28 November 2013).
Gunderson, C. A., N. T. Edwards, A. V. Walker, K. H. O'Hara, C. M. Campion, and P. J.
Hanson. 2012. Forest phenology and a warmer climate–growing season extension in
relation to climatic provenance. Global Change Biology 18(6): 2008-2025.
Hall, M. H. P., and D. B. Fagre. 2003. Modeled Climate-Induced Glacier Change in Glacier
National Park, 1850-2100. BioScience 53(2): 131-140.
Hansen, A. J., R. P. Neilson, V. H. Dale, C. H. Flather, L. R. Iverson, D. J. Currie, S. Shafer, R.
Cook, and P. J. Bartlein. 2001. Global Change in Forests: Responses of Species,
Communities, and Biomes. BioScience 51(9): 765-779.
Hanson, P. J., and J. F. Weltzin. 2000. Drought disturbance from climate change: response of
United States forests. The Science of the Total Environment 262(3): 205-220.
Hill, J., and H. M. Mogil. 2012. The Weather and Climate of Kentucky. Weatherwise 65(5): 2632.
Houghton, J. T., Y. Ding, D. J. Griggs, M. Noguer, P. J. van der Linden, X. Dai, K. Maskell, and
C. A. Johnson. 2001. Climate Change 2001: The Scientific Basis. New York: Cambridge
University Press.
Huete, A., C. Justice, and W. van Leeuwen. 1999. MODIS Vegetation Index (MOD 13)
Algorithm Theoretical Basis Document. (ATBD) Version 3.0. Greenbelt, MD: EOS
Project Office, NASA Goddard Space Flight Center.
71
Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of
the radiometric and biophysical performance of the MODIS vegetation indices. Remote
Sensing of Environment: 83(1): 195–213.
Hufkens, K., M. Friedl, O. Sonnentag, B. H. Braswell, T. Milliman, and A. D. Richardson. 2012.
Linking near-surface and satellite remote sensing measurements of deciduous broadleaf
forest phenology. Remote Sensing of Environment 117(15): 307-321.
Hurrell, J. W. 1995. Decadal Trends in the North Atlantic Oscillation: Regional Temperatures
and Precipitation. Science 269(5224): 676-679.
Hurrell, J. W., and H. van Loon. 1997. Decadal variations in climate associated with the North
Atlantic Oscillation. Climate Change 36(3-4): 301-326.
Irland, L. C., D. Adams, R. Alig, C. J. Betz, C. Chen, M. Hutchins, B. A. McCarl, K. Skog, and
B. L. Sohngen. 2001. Assessing Socioeconomic Impacts of Climate Change on US
Forests, Wood-Product Markets, and Forest Recreation. BioScience 51(9): 753-764.
Iverson, L. R., and A. M. Prasad. 2001. Potential changes in tree species richness and forest
community types following climate change. Ecosystems 4(3): 186-199.
Jennings, M. D., D. Faber-Langendoen, R. K. Peet, O. L. Loucks, D. C. Glenn-Lewin, A.
Damman, M. G. Barbour, R. Pfister, D. H. Grossman, D. Roberts, D. Tart, M. Walker, S.
S. Talbot, J. Walker, G. S. Hartshorn, G. Waggoner, M. D. Abrams, A. Hill, and M.
Rejmanek. 2006. Description, Documentation, And Evaluation Of Associations And
Alliances Within The U.S. National Vegetation Classification, Version 4.5. Washington
DC: Ecological Society of America, Vegetation Classification Panel.
JISAO (Joint Institute for the Study of the Atmosphere and Ocean). 2013. PDO Index.
University of Washington, College of the Environment. Available at
http://jisao.washington.edu/pdo/PDO.latest (last accessed 8 November 2013).
Jordan, T. R., and M. Madden. 2010. Digital Vegetation Maps for the NPS CumberlandPiedmont I&M Network. Natural Resource Technical Report. Fort Collins, CO: National
Park Service.
Justice, C. O., E. Vermote, J. R. G. Townshend, R. Defries, D. P. Roy, D. K. Hall,
V. V. Salomonson, J. L. Privette, G. Riggs, A. Strahler, W. Lucht, R. P. Myneni, Y.
Knyazikhin, S. W. Running, R. R. Nemani, Z. M. Wan, A. R. Huete, W. van
Leeuwen, R. E. Wolfe, L. Giglio, J-P Muller, P. Lewis, and M. J. Barnsley.
1998. The moderate resolution imaging spectroradiometer (MODIS): land remote sensing
for global change research. IEEE Transactions on Geoscience and Remote Sensing 36(4):
1228 -1249.
72
Kennedy, M. C., and P. J. Watson. 1997. The Chronology of Early Agriculture and Intensive
Mineral Mining in the Salts Cave and Mammoth Cave Region, Mammoth Cave National
Park, Kentucky. Journal of Caves and Karst Studies 59: 5-9.
Kentucky Geological Survey. 2002. Geology of Kentucky. Lexington, KY: Kentucky Geological
Survey. Available at http://kgs.uky.edu/kgsmap/kgsgeoserver/viewer.asp (last accessed
27 November 2013).
Kerr, J. T., and M. Ostrovsky. 2003. From space to species: ecological applications for remote
sensing. Trends in Ecology and Evolution 18(6): 299-305.
Kiers, E. T., T. M. Palmer, A. R. Ives, J. F. Bruno, and J. L. Bronstein. 2010. Mutualisms in a
changing world: an evolutionary perspective. Ecology Letters 13(12): 1459-1474.
Kukowski, K. R., S. Schwinning, and B. F. Schwartz. 2013. Hydraulic responses to extreme
drought conditions in three co-dominant tree species in shallow soil over
bedrock. Oecologia 171(4): 1-12.
Kurtzman, D., and B. R. Scanlon. 2007. El Niño–Southern Oscillation and Pacific Decadal
Oscillation impacts on precipitation in the southern and central United States: Evaluation
of spatial distribution and predictions. Water Resources Research 43(10): W10427.
LP DAAC (Land Processes Distributed Active Archive Center). 2013. LP DAAC Data Pool.
U.S. Geological Survey. Available at https://lpdaac.usgs.gov/data_access/data_pool (last
accessed 13 November 2013).
Leibfreid, T. R., R. L. Woodman, and S. C. Thomas. 2005. Vital Signs Monitoring Plan for the
Cumberland Piedmont Network and Mammoth Cave National Park Prototype
Monitoring Program: July 2005. Mammoth Cave, Kentucky: National Park Service.
Leathers, D. J., B. Yarnal, and M. A. Palecki. 1991. The Pacific/North American Teleconnection
Pattern and United States Climate. Part I: Regional Temperature and Precipitation
Associations. Journal of Climate 4(5): 517-528.
Leathers, D. J., and M. A. Palecki. 1992. The Pacific/North American Teleconnection Pattern
and United States Climate. Part II: Temporal Characteristics and Index Specification.
Journal of Climate 5(7): 707-716.
Maselli, F. 2004. Monitoring forest conditions in a protected Mediterranean coastal area by the
analysis of multiyear NDVI data. Remote sensing of Environment 89(4): 423-433.
Mantua, N. J., and S. R. Hare. 2002. The Pacific Decadal Oscillation. Journal of Oceanography,
58(1): 35-44.
73
McEwan, R. W., R. D. Paratley, R. N. Muller, and C. L. Riccardi. 2005. The vascular flora of an
old-growth mixed mesophytic forest in southeastern Kentucky. The Journal of the Torrey
Botanical Society 132(4): 618–627.
McNulty, S. G., and J. D. Aber. 2001. US National Climate Change Assessment of Forest
Ecosystems: An Introduction. BioScience 51(9): 720-722.
NASA. 2013. Measuring Vegetation (NDVI & EVI). Greenbelt, MD: Earth Observatory, NASA
Goddard Space Flight Center. Available at
http://earthobservatory.nasa.gov/Features/MeasuringVegetation/measuring_vegetation_2.
php (last accessed 4 November 2013).
NCDC (National Climatic Data Center). 2013a. Climate Data Online. National Oceanic and
Atmospheric Administration, National Climatic Data Center. Available at
http://www.ncdc.noaa.gov/cdo-web/ (last accessed 18 November 2013).
———. 2013b. NOAA/NESDIS North American Drought Variability. National Oceanic and
Atmospheric Administration, NOAA Paleoclimatology. Available at
http://www.ncdc.noaa.gov/paleo/pdsi.html (last accessed 10 November 2013).
———. 2013c. Climatological Rankings. National Oceanic and Atmospheric Administration,
National Climatic Data Center. Available at http://www.ncdc.noaa.gov/temp-andprecip/ranks.php (last accessed 30 November 2013).
NPS (National Park Service). 2013. Current Administrative Boundaries of National Park System
Units. Geospatial Dataset- 2203224. Ft. Collins, CO: National Park Service Land
Resources Division..
Natural Earth. 2013. 1:10m Cultural Vectors. Available at
http://www.naturalearthdata.com/downloads/10m-cultural-vectors/ (last accessed 18
November 2013).
Palmer, A. N. 1981. A geological guide to Mammoth Cave National Park. Teaneck, NJ:
Zephyrus Press.
———. 1991. Origin and morphology of limestone caves. Geological Society of America
Bulletin 103(1): 1-21.
Pearson, R. G., and T. P. Dawson. 2003. Predicting the impacts of climate change on the
distribution of species: are bioclimate envelope models useful? Global Ecology &
Biogeography 12(5): 361-371.
74
Pelkey, N. W., C. J. Stoner, and T. M. Caro. 2000. Vegetation in Tanzania: assessing long term
trends and effects of protection using satellite imagery. Biological Conservation 94(3):
297-309.
Pettorelli, N., J. O. Vik, A. Mysterud, J. Gaillard, C. J. Tucker, and N. C. Stenseth. 2005. Using
the Satellite-derived NDVI to Assess Ecological Responses to Environmental Change.
Trends in Ecology & Evolution 20(9): 503-510.
Polgar, C. A., and R. B. Primack. 2011. Leaf‐out phenology of temperate woody plants: from
trees to ecosystems. New Phytologist 191(4): 926-941.
Pouliot, D., R. Latifovic, and I. Olthof. 2009. Trends in vegetation NDVI from 1 km AVHRR
data over Canada for the period 1985–2006. International Journal of Remote
Sensing 30(1): 149-168.
Rodwell, M. J., D. P. Rowell, and C. K. Folland. 1999. Oceanic forcing of the wintertime North
Atlantic Oscillation and European climate. Nature 398(6725): 320-323.
Seager, R., A. Tzanova, and J. Nakamura. 2009. Drought in the Southeastern United States:
Causes, Variability over the Last Millennium, and the Potential for Future Hydroclimate
Change. Journal of Climate 22(19): 5021- 5045.
Seager, R., Y. Kushnir, J. Nakamura, M. Ting, and N. Naik. 2010. Northern Hemisphere winter
snow anomalies: ENSO, NAO and the winter of 2009/10. Geophysical Research
Letters 37(14): DOI: 10.1029/2010GL043830.
Smart, L., and R. White. 2010. Accuracy Assessment: Mammoth Cave National Park. Durham,
NC: NatureServe.
Smith, S. J., A. M. Thomson, N. J. Rosenberg, R. C. Izaurralde, R. A. Brown, and T. M. L.
Wigley. 2005. Climate Change Impacts for the Conterminous USA: An Integrated
Assessment. Climatic Change 69(1): 7-25.
Tang, Z., J. Fang, J. Sun, and K. J. Gaston. 2011. Effectiveness of protected areas in maintaining
plant production. PLoS ONE 6(4): DOI: 10.1371/journal.pone.0019116.
Thompson, I., B. Mackey, S. McNulty, and A. Mosseler. 2009. Forest resilience, biodiversity,
and climate change. A synthesis of the biodiversity/resilience/stability relation in forest
ecosystems. Montreal: Secretariat of the Convention on Biological Diversity.
Turner, W., S. Spector, N. Gardiner, M. Fladeland, E. Sterling, and M. Steininger. 2003. Remote
sensing for biodiversity science and conservation. Trends in Ecology & Evolution 18(6):
306-314.
75
Tylianakis, J. M., R. K. Didham, J. Bascompte, and D. A. Wardle. 2008. Global change and
species interactions in terrestrial ecosystems. Ecology Letters 11(12): 1351-1363.
Vowels, K. M. 2012. Ice storm damage to upland oak-hickory forest at Bernheim Forest,
Kentucky. The Journal of the Torrey Botanical Society 139(4): 406-415.
Wallace, J. M., and D. S. Gutzler. 1981. Teleconnections in the geopotential height field during
the Northern Hemisphere winter. Monthly Weather Review 109(4): 784-812.
Walther, G-R., E. Post, P. Convey, A. Menzel, C. Parmesan, T. Beebee, J-M. Fromentin, O.
Hoegh-Guldberg, and F. Bairlein. 2002. Ecological responses to recent climate change.
Nature 416(6879): 389–395.
Wang, J., P. M. Rich, and K. P. Price. 2003. Temporal responses of NDVI to precipitation and
temperature in the central Great Plains, USA. International Journal of Remote Sensing
24(11): 2345-2364.
Wanner, H., S. Brönnimann, C. Casty, D. Gyalistras, J. Luterbacher, C. Schmutz, D. B.
Stephenson, and E. Xoplaki. 2001. North Atlantic Oscillation–concepts and
studies. Surveys in Geophysics 22(4): 321-381.
Warmund, M. R., P. Guinan, and G. Fernandez. 2008. Temperatures and cold damage to small
fruit crops across the eastern United States associated with the April 2007
freeze. HortScience 43(6): 1643-1647.
White, W. B., R. A. Watson, E. R. Pohl, and R. Brucker. 1970. The central Kentucky
karst. Geographical Review 60(1): 88-115.
Wolter, K., and M. S. Timlin. 1993. Monitoring ENSO in COADS with a seasonally adjusted
principal component index. Proceedings of the 17th Climate Diagnostics Workshop 5257.
———. 2011. El Niño/Southern Oscillation behavior since 1871 as diagnosed in an extended
multivariate ENSO index (MEI.ext). International Journal of Climatology 31(7): 10741087.
Wu, A., W. W. Hsieh, and A. Shabbar. 2005. The Nonlinear Patterns of North American Winter
Temperature and Precipitation Associated with ENSO. Journal of Climate 18(11): 17361752.
Zhao, M., and S. W. Running. 2010. Drought-induced reduction in global terrestrial net primary
production from 2000 through 2009. Science 329(5994): 940-943.
76